{ "name": "ModelVRAM LLM VRAM requirements", "version": "1.2026-09-29b", "schema_version": 1, "source_checked": "2026-09-29", "url": "https://modelvram.com/data/", "csv": "https://modelvram.com/data/modelvram-vram.csv", "license": { "name": "CC BY 4.0", "url": "https://creativecommons.org/licenses/by/4.0/" }, "attribution": "Data: ModelVRAM (modelvram.com) https://modelvram.com/data/", "units": { "gib": "GiB (1024^3 bytes); the site writes it as GB", "params": "parameters", "tokens": "tokens" }, "assumptions": { "requests": 1, "kv_cache": "FP16", "overhead": "0.5 GiB + 10% of weights and KV cache", "bits_per_weight": { "q4_k_m": 4.84, "q8_0": 8.5, "fp16": 16 } }, "columns": [ { "name": "model", "unit": "", "description": "Model name as shown on the site" }, { "name": "hf_id", "unit": "", "description": "Hugging Face repository ID" }, { "name": "hf_url", "unit": "URL", "description": "Source repository on Hugging Face" }, { "name": "model_url", "unit": "URL", "description": "ModelVRAM page with the full calculations" }, { "name": "source_checked", "unit": "date", "description": "When config.json and file sizes were read" }, { "name": "total_params", "unit": "parameters", "description": "All parameters, every expert included" }, { "name": "active_params", "unit": "parameters", "description": "Parameters used per token (MoE); empty for dense models" }, { "name": "experts", "unit": "count", "description": "Routed experts (MoE); empty for dense models" }, { "name": "layers", "unit": "count", "description": "Transformer layers" }, { "name": "max_context", "unit": "tokens", "description": "Longest context in the model config" }, { "name": "kv_bytes_per_token_fp16", "unit": "bytes", "description": "KV cache added per token of one request at FP16, once sliding windows are full" }, { "name": "attention_layout", "unit": "", "description": "Layers that decide the KV cache" }, { "name": "q4_k_m_weights_gib", "unit": "GiB", "description": "Weight memory at Q4_K_M" }, { "name": "q8_0_weights_gib", "unit": "GiB", "description": "Weight memory at Q8_0" }, { "name": "fp16_weights_gib", "unit": "GiB", "description": "Weight memory at FP16 / BF16" }, { "name": "q4_k_m_8k_total_gib", "unit": "GiB", "description": "Total VRAM at Q4_K_M, 8,192 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "q4_k_m_32k_total_gib", "unit": "GiB", "description": "Total VRAM at Q4_K_M, 32,768 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "q4_k_m_128k_total_gib", "unit": "GiB", "description": "Total VRAM at Q4_K_M, 131,072 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "q8_0_8k_total_gib", "unit": "GiB", "description": "Total VRAM at Q8_0, 8,192 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "q8_0_32k_total_gib", "unit": "GiB", "description": "Total VRAM at Q8_0, 32,768 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "q8_0_128k_total_gib", "unit": "GiB", "description": "Total VRAM at Q8_0, 131,072 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "fp16_8k_total_gib", "unit": "GiB", "description": "Total VRAM at FP16 / BF16, 8,192 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "fp16_32k_total_gib", "unit": "GiB", "description": "Total VRAM at FP16 / BF16, 32,768 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "fp16_128k_total_gib", "unit": "GiB", "description": "Total VRAM at FP16 / BF16, 131,072 tokens: weights + KV cache + overhead; empty past max_context" }, { "name": "kv_cache_8k_gib", "unit": "GiB", "description": "FP16 KV cache of one request at 8,192 tokens" }, { "name": "kv_cache_32k_gib", "unit": "GiB", "description": "FP16 KV cache of one request at 32,768 tokens" }, { "name": "kv_cache_128k_gib", "unit": "GiB", "description": "FP16 KV cache of one request at 131,072 tokens" }, { "name": "q4_k_m_32k_fits_16gib", "unit": "boolean", "description": "Q4_K_M with 32,768 tokens fits on one 16 GB GPU" }, { "name": "q4_k_m_32k_fits_24gib", "unit": "boolean", "description": "Q4_K_M with 32,768 tokens fits on one 24 GB GPU" }, { "name": "q4_k_m_32k_fits_32gib", "unit": "boolean", "description": "Q4_K_M with 32,768 tokens fits on one 32 GB GPU" }, { "name": "q4_k_m_32k_fits_48gib", "unit": "boolean", "description": "Q4_K_M with 32,768 tokens fits on one 48 GB GPU" }, { "name": "q4_k_m_32k_fits_80gib", "unit": "boolean", "description": "Q4_K_M with 32,768 tokens fits on one 80 GB GPU" }, { "name": "q8_0_32k_fits_16gib", "unit": "boolean", "description": "Q8_0 with 32,768 tokens fits on one 16 GB GPU" }, { "name": "q8_0_32k_fits_24gib", "unit": "boolean", "description": "Q8_0 with 32,768 tokens fits on one 24 GB GPU" }, { "name": "q8_0_32k_fits_32gib", "unit": "boolean", "description": "Q8_0 with 32,768 tokens fits on one 32 GB GPU" }, { "name": "q8_0_32k_fits_48gib", "unit": "boolean", "description": "Q8_0 with 32,768 tokens fits on one 48 GB GPU" }, { "name": "q8_0_32k_fits_80gib", "unit": "boolean", "description": "Q8_0 with 32,768 tokens fits on one 80 GB GPU" }, { "name": "fp16_32k_fits_16gib", "unit": "boolean", "description": "FP16 / BF16 with 32,768 tokens fits on one 16 GB GPU" }, { "name": "fp16_32k_fits_24gib", "unit": "boolean", "description": "FP16 / BF16 with 32,768 tokens fits on one 24 GB GPU" }, { "name": "fp16_32k_fits_32gib", "unit": "boolean", "description": "FP16 / BF16 with 32,768 tokens fits on one 32 GB GPU" }, { "name": "fp16_32k_fits_48gib", "unit": "boolean", "description": "FP16 / BF16 with 32,768 tokens fits on one 48 GB GPU" }, { "name": "fp16_32k_fits_80gib", "unit": "boolean", "description": "FP16 / BF16 with 32,768 tokens fits on one 80 GB GPU" }, { "name": "note", "unit": "", "description": "Known cache optimization the estimate leaves out" } ], "models": [ { "model": "Spark-X2.5 4B", "hf_id": "XHToken/Spark-X2.5-4B", "hf_url": "https://huggingface.co/XHToken/Spark-X2.5-4B", "model_url": "https://modelvram.com/llm-vram-calculator/spark-x2.5-4b/", "source_checked": "2026-09-29", "total_params": 4112079360, "active_params": null, "experts": null, "layers": 36, "max_context": 1048576, "kv_bytes_per_token_fp16": 36864, "attention_layout": "9 of its 36 layers use full attention and 27 keep a sliding window of 512 tokens", "q4_k_m_weights_gib": 2.32, "q8_0_weights_gib": 4.07, "fp16_weights_gib": 7.66, "q4_k_m_8k_total_gib": 3.47, "q4_k_m_32k_total_gib": 4.4, "q4_k_m_128k_total_gib": 8.11, "q8_0_8k_total_gib": 5.4, "q8_0_32k_total_gib": 6.33, "q8_0_128k_total_gib": 10.04, "fp16_8k_total_gib": 9.35, "fp16_32k_total_gib": 10.28, "fp16_128k_total_gib": 13.99, "kv_cache_8k_gib": 0.39, "kv_cache_32k_gib": 1.23, "kv_cache_128k_gib": 4.61, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": true, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Ling 3.0 Tiny 7.9B-A1.3B (MoE)", "hf_id": "inclusionAI/Ling-3.0-tiny", "hf_url": "https://huggingface.co/inclusionAI/Ling-3.0-tiny", "model_url": "https://modelvram.com/llm-vram-calculator/ling-3.0-tiny/", "source_checked": "2026-09-29", "total_params": 7893392800, "active_params": 1300000000, "experts": 128, "layers": 24, "max_context": 131072, "kv_bytes_per_token_fp16": 6912, "attention_layout": "6 of its 24 layers use multi-head latent attention and 18 are KDA recurrent layers with no growing cache", "q4_k_m_weights_gib": 4.45, "q8_0_weights_gib": 7.81, "fp16_weights_gib": 14.7, "q4_k_m_8k_total_gib": 5.45, "q4_k_m_32k_total_gib": 5.62, "q4_k_m_128k_total_gib": 6.32, "q8_0_8k_total_gib": 9.15, "q8_0_32k_total_gib": 9.32, "q8_0_128k_total_gib": 10.02, "fp16_8k_total_gib": 16.73, "fp16_32k_total_gib": 16.9, "fp16_128k_total_gib": 17.6, "kv_cache_8k_gib": 0.05, "kv_cache_32k_gib": 0.21, "kv_cache_128k_gib": 0.84, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "The 18 KDA layers keep a fixed recurrent state, which this estimate does not include; only the 6 MLA layers add a context-growing KV cache. Real serving memory may be higher." }, { "model": "LFM2.5 8B-A1B (MoE)", "hf_id": "LiquidAI/LFM2.5-8B-A1B", "hf_url": "https://huggingface.co/LiquidAI/LFM2.5-8B-A1B", "model_url": "https://modelvram.com/llm-vram-calculator/lfm2.5-8b-a1b/", "source_checked": "2026-09-29", "total_params": 8467856832, "active_params": 1500000000, "experts": 32, "layers": 24, "max_context": 128000, "kv_bytes_per_token_fp16": 12288, "attention_layout": "6 of its 24 layers use full attention and 18 are short-convolution layers with no growing cache", "q4_k_m_weights_gib": 4.77, "q8_0_weights_gib": 8.38, "fp16_weights_gib": 15.77, "q4_k_m_8k_total_gib": 5.85, "q4_k_m_32k_total_gib": 6.16, "q4_k_m_128k_total_gib": null, "q8_0_8k_total_gib": 9.82, "q8_0_32k_total_gib": 10.13, "q8_0_128k_total_gib": null, "fp16_8k_total_gib": 17.95, "fp16_32k_total_gib": 18.26, "fp16_128k_total_gib": null, "kv_cache_8k_gib": 0.09, "kv_cache_32k_gib": 0.38, "kv_cache_128k_gib": null, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Step 3.7 Flash 196B-A11B (MoE)", "hf_id": "stepfun-ai/Step-3.7-Flash", "hf_url": "https://huggingface.co/stepfun-ai/Step-3.7-Flash", "model_url": "https://modelvram.com/llm-vram-calculator/step-3.7-flash/", "source_checked": "2026-09-29", "total_params": 201365316160, "active_params": 11000000000, "experts": 288, "layers": 45, "max_context": 262144, "kv_bytes_per_token_fp16": 49152, "attention_layout": "12 of its 45 layers use full attention and 33 keep a sliding window of 512 tokens", "q4_k_m_weights_gib": 113.46, "q8_0_weights_gib": 199.26, "fp16_weights_gib": 375.07, "q4_k_m_8k_total_gib": 125.86, "q4_k_m_32k_total_gib": 127.1, "q4_k_m_128k_total_gib": 132.05, "q8_0_8k_total_gib": 220.24, "q8_0_32k_total_gib": 221.47, "q8_0_128k_total_gib": 226.42, "fp16_8k_total_gib": 413.63, "fp16_32k_total_gib": 414.87, "fp16_128k_total_gib": 419.82, "kv_cache_8k_gib": 0.5, "kv_cache_32k_gib": 1.63, "kv_cache_128k_gib": 6.13, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Hy4 Preview 770B-A49B (MoE)", "hf_id": "tencent/Hy4-preview", "hf_url": "https://huggingface.co/tencent/Hy4-preview", "model_url": "https://modelvram.com/llm-vram-calculator/hy4-preview/", "source_checked": "2026-09-29", "total_params": 779960992733, "active_params": 49000000000, "experts": 256, "layers": 78, "max_context": 1048576, "kv_bytes_per_token_fp16": 92544, "attention_layout": "All 78 layers use multi-head latent attention", "q4_k_m_weights_gib": 439.47, "q8_0_weights_gib": 771.79, "fp16_weights_gib": 1452.79, "q4_k_m_8k_total_gib": 484.69, "q4_k_m_32k_total_gib": 487.02, "q4_k_m_128k_total_gib": 496.34, "q8_0_8k_total_gib": 850.25, "q8_0_32k_total_gib": 852.58, "q8_0_128k_total_gib": 861.9, "fp16_8k_total_gib": 1599.35, "fp16_32k_total_gib": 1601.68, "fp16_128k_total_gib": 1611, "kv_cache_8k_gib": 0.71, "kv_cache_32k_gib": 2.82, "kv_cache_128k_gib": 11.3, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Limite 1B Violetto", "hf_id": "paradigma-inc/limite-1b-violetto", "hf_url": "https://huggingface.co/paradigma-inc/limite-1b-violetto", "model_url": "https://modelvram.com/llm-vram-calculator/limite-1b-violetto/", "source_checked": "2026-09-29", "total_params": 1035253888, "active_params": null, "experts": null, "layers": 48, "max_context": 131072, "kv_bytes_per_token_fp16": 12288, "attention_layout": "12 of its 48 layers use full attention and 36 keep a sliding window of 1,025 tokens", "q4_k_m_weights_gib": 0.58, "q8_0_weights_gib": 1.02, "fp16_weights_gib": 1.93, "q4_k_m_8k_total_gib": 1.31, "q4_k_m_32k_total_gib": 1.62, "q4_k_m_128k_total_gib": 2.86, "q8_0_8k_total_gib": 1.8, "q8_0_32k_total_gib": 2.11, "q8_0_128k_total_gib": 3.34, "fp16_8k_total_gib": 2.79, "fp16_32k_total_gib": 3.1, "fp16_128k_total_gib": 4.34, "kv_cache_8k_gib": 0.16, "kv_cache_32k_gib": 0.44, "kv_cache_128k_gib": 1.56, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": true, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "ZDTaichu 5.0 9B", "hf_id": "TaichuAI/ZDTaichu5.0-9B", "hf_url": "https://huggingface.co/TaichuAI/ZDTaichu5.0-9B", "model_url": "https://modelvram.com/llm-vram-calculator/zdtaichu5.0-9b/", "source_checked": "2026-09-29", "total_params": 9794197512, "active_params": null, "experts": null, "layers": 32, "max_context": 131072, "kv_bytes_per_token_fp16": 32768, "attention_layout": "8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 5.52, "q8_0_weights_gib": 9.69, "fp16_weights_gib": 18.24, "q4_k_m_8k_total_gib": 6.85, "q4_k_m_32k_total_gib": 7.67, "q4_k_m_128k_total_gib": 10.97, "q8_0_8k_total_gib": 11.44, "q8_0_32k_total_gib": 12.26, "q8_0_128k_total_gib": 15.56, "fp16_8k_total_gib": 20.84, "fp16_32k_total_gib": 21.67, "fp16_128k_total_gib": 24.97, "kv_cache_8k_gib": 0.25, "kv_cache_32k_gib": 1, "kv_cache_128k_gib": 4, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Hemmingway-1 27B", "hf_id": "Altworld/Hemmingway-1", "hf_url": "https://huggingface.co/Altworld/Hemmingway-1", "model_url": "https://modelvram.com/llm-vram-calculator/hemmingway-1/", "source_checked": "2026-09-29", "total_params": 27320697856, "active_params": null, "experts": null, "layers": 64, "max_context": 262144, "kv_bytes_per_token_fp16": 65536, "attention_layout": "16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 15.39, "q8_0_weights_gib": 27.03, "fp16_weights_gib": 50.89, "q4_k_m_8k_total_gib": 17.98, "q4_k_m_32k_total_gib": 19.63, "q4_k_m_128k_total_gib": 26.23, "q8_0_8k_total_gib": 30.79, "q8_0_32k_total_gib": 32.44, "q8_0_128k_total_gib": 39.04, "fp16_8k_total_gib": 57.03, "fp16_32k_total_gib": 58.68, "fp16_128k_total_gib": 65.28, "kv_cache_8k_gib": 0.5, "kv_cache_32k_gib": 2, "kv_cache_128k_gib": 8, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "MiMo V2.6 Distill Qwen 9B", "hf_id": "XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B", "hf_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Distill-Qwen-9B", "model_url": "https://modelvram.com/llm-vram-calculator/mimo-v2.6-distill-qwen-9b/", "source_checked": "2026-09-29", "total_params": 9409813744, "active_params": null, "experts": null, "layers": 32, "max_context": 262144, "kv_bytes_per_token_fp16": 32768, "attention_layout": "8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 5.3, "q8_0_weights_gib": 9.31, "fp16_weights_gib": 17.53, "q4_k_m_8k_total_gib": 6.61, "q4_k_m_32k_total_gib": 7.43, "q4_k_m_128k_total_gib": 10.73, "q8_0_8k_total_gib": 11.02, "q8_0_32k_total_gib": 11.84, "q8_0_128k_total_gib": 15.14, "fp16_8k_total_gib": 20.05, "fp16_32k_total_gib": 20.88, "fp16_128k_total_gib": 24.18, "kv_cache_8k_gib": 0.25, "kv_cache_32k_gib": 1, "kv_cache_128k_gib": 4, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "K2-Horizon MoVA 36B-A4B (MoE)", "hf_id": "IFM/K2-Horizon-MoVA-36B-A4B", "hf_url": "https://huggingface.co/IFM/K2-Horizon-MoVA-36B-A4B", "model_url": "https://modelvram.com/llm-vram-calculator/k2-horizon-mova-36b-a4b/", "source_checked": "2026-09-29", "total_params": 37444792020, "active_params": 4000000000, "experts": 100, "layers": 48, "max_context": 524288, "kv_bytes_per_token_fp16": 196608, "attention_layout": "All 48 layers use full attention", "q4_k_m_weights_gib": 21.1, "q8_0_weights_gib": 37.05, "fp16_weights_gib": 69.75, "q4_k_m_8k_total_gib": 25.36, "q4_k_m_32k_total_gib": 30.31, "q4_k_m_128k_total_gib": 50.11, "q8_0_8k_total_gib": 42.91, "q8_0_32k_total_gib": 47.86, "q8_0_128k_total_gib": 67.66, "fp16_8k_total_gib": 78.87, "fp16_32k_total_gib": 83.82, "fp16_128k_total_gib": 103.62, "kv_cache_8k_gib": 1.5, "kv_cache_32k_gib": 6, "kv_cache_128k_gib": 24, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "IQuest-Q1 320B-A15B (MoE)", "hf_id": "IQuestLab/IQuest-Q1", "hf_url": "https://huggingface.co/IQuestLab/IQuest-Q1", "model_url": "https://modelvram.com/llm-vram-calculator/iquest-q1/", "source_checked": "2026-09-29", "total_params": 320318615552, "active_params": 15000000000, "experts": 256, "layers": 88, "max_context": 524288, "kv_bytes_per_token_fp16": 102400, "attention_layout": "25 of its 88 layers use full attention and 63 keep a sliding window of 4,096 tokens", "q4_k_m_weights_gib": 180.48, "q8_0_weights_gib": 316.96, "fp16_weights_gib": 596.64, "q4_k_m_8k_total_gib": 201.11, "q4_k_m_32k_total_gib": 203.69, "q4_k_m_128k_total_gib": 214, "q8_0_8k_total_gib": 351.24, "q8_0_32k_total_gib": 353.82, "q8_0_128k_total_gib": 364.13, "fp16_8k_total_gib": 658.88, "fp16_32k_total_gib": 661.46, "fp16_128k_total_gib": 671.77, "kv_cache_8k_gib": 1.89, "kv_cache_32k_gib": 4.23, "kv_cache_128k_gib": 13.61, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "LLM-jp-4.1 32B-A3B Thinking (MoE)", "hf_id": "llm-jp/llm-jp-4.1-32b-a3b-thinking", "hf_url": "https://huggingface.co/llm-jp/llm-jp-4.1-32b-a3b-thinking", "model_url": "https://modelvram.com/llm-vram-calculator/llm-jp-4.1-32b-a3b-thinking/", "source_checked": "2026-09-29", "total_params": 32139028992, "active_params": 3827476992, "experts": 128, "layers": 32, "max_context": 65536, "kv_bytes_per_token_fp16": 65536, "attention_layout": "All 32 layers use full attention", "q4_k_m_weights_gib": 18.11, "q8_0_weights_gib": 31.8, "fp16_weights_gib": 59.86, "q4_k_m_8k_total_gib": 20.97, "q4_k_m_32k_total_gib": 22.62, "q4_k_m_128k_total_gib": null, "q8_0_8k_total_gib": 36.03, "q8_0_32k_total_gib": 37.68, "q8_0_128k_total_gib": null, "fp16_8k_total_gib": 66.9, "fp16_32k_total_gib": 68.55, "fp16_128k_total_gib": null, "kv_cache_8k_gib": 0.5, "kv_cache_32k_gib": 2, "kv_cache_128k_gib": null, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Ornith 1.5 35B-A3B (MoE)", "hf_id": "ornith-ai/Ornith-1.5-35B-A3B", "hf_url": "https://huggingface.co/ornith-ai/Ornith-1.5-35B-A3B", "model_url": "https://modelvram.com/llm-vram-calculator/ornith-1.5-35b-a3b/", "source_checked": "2026-09-26", "total_params": 35951822704, "active_params": 3000000000, "experts": 256, "layers": 40, "max_context": 262144, "kv_bytes_per_token_fp16": 20480, "attention_layout": "10 of its 40 layers use full attention and 30 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 20.26, "q8_0_weights_gib": 35.58, "fp16_weights_gib": 66.97, "q4_k_m_8k_total_gib": 22.95, "q4_k_m_32k_total_gib": 23.47, "q4_k_m_128k_total_gib": 25.53, "q8_0_8k_total_gib": 39.8, "q8_0_32k_total_gib": 40.32, "q8_0_128k_total_gib": 42.38, "fp16_8k_total_gib": 74.33, "fp16_32k_total_gib": 74.85, "fp16_128k_total_gib": 76.91, "kv_cache_8k_gib": 0.16, "kv_cache_32k_gib": 0.63, "kv_cache_128k_gib": 2.5, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Ornith 1.5 9B", "hf_id": "ornith-ai/Ornith-1.5-9B", "hf_url": "https://huggingface.co/ornith-ai/Ornith-1.5-9B", "model_url": "https://modelvram.com/llm-vram-calculator/ornith-1.5-9b/", "source_checked": "2026-09-26", "total_params": 9653104368, "active_params": null, "experts": null, "layers": 32, "max_context": 262144, "kv_bytes_per_token_fp16": 32768, "attention_layout": "8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 5.44, "q8_0_weights_gib": 9.55, "fp16_weights_gib": 17.98, "q4_k_m_8k_total_gib": 6.76, "q4_k_m_32k_total_gib": 7.58, "q4_k_m_128k_total_gib": 10.88, "q8_0_8k_total_gib": 11.28, "q8_0_32k_total_gib": 12.11, "q8_0_128k_total_gib": 15.41, "fp16_8k_total_gib": 20.55, "fp16_32k_total_gib": 21.38, "fp16_128k_total_gib": 24.68, "kv_cache_8k_gib": 0.25, "kv_cache_32k_gib": 1, "kv_cache_128k_gib": 4, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "AliceAI Foundation 80B-A3B (MoE)", "hf_id": "yandex/AliceAI-Foundation-80B-A3B-Base", "hf_url": "https://huggingface.co/yandex/AliceAI-Foundation-80B-A3B-Base", "model_url": "https://modelvram.com/llm-vram-calculator/aliceai-foundation-80b-a3b-base/", "source_checked": "2026-09-27", "total_params": 81286433408, "active_params": 3000000000, "experts": 512, "layers": 48, "max_context": 262144, "kv_bytes_per_token_fp16": 24576, "attention_layout": "12 of its 48 layers use full attention and 36 are KDA recurrent layers with no growing cache", "q4_k_m_weights_gib": 45.8, "q8_0_weights_gib": 80.44, "fp16_weights_gib": 151.41, "q4_k_m_8k_total_gib": 51.09, "q4_k_m_32k_total_gib": 51.71, "q4_k_m_128k_total_gib": 54.18, "q8_0_8k_total_gib": 89.19, "q8_0_32k_total_gib": 89.8, "q8_0_128k_total_gib": 92.28, "fp16_8k_total_gib": 167.25, "fp16_32k_total_gib": 167.87, "fp16_128k_total_gib": 170.35, "kv_cache_8k_gib": 0.19, "kv_cache_32k_gib": 0.75, "kv_cache_128k_gib": 3, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "The 36 KDA layers keep a fixed recurrent state, which this estimate does not include; only the 12 gated-attention layers add a context-growing KV cache. Real serving memory may be higher." }, { "model": "DeepSeek V4 Flash 0731", "hf_id": "deepseek-ai/DeepSeek-V4-Flash-0731", "hf_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731", "model_url": "https://modelvram.com/llm-vram-calculator/deepseek-v4-flash-0731/", "source_checked": "2026-09-26", "total_params": 304180418494, "active_params": 13000000000, "experts": 256, "layers": 43, "max_context": 1048576, "kv_bytes_per_token_fp16": 88064, "attention_layout": "All 43 layers use full attention", "q4_k_m_weights_gib": 171.39, "q8_0_weights_gib": 301, "fp16_weights_gib": 566.58, "q4_k_m_8k_total_gib": 189.77, "q4_k_m_32k_total_gib": 191.99, "q4_k_m_128k_total_gib": 200.85, "q8_0_8k_total_gib": 332.33, "q8_0_32k_total_gib": 334.55, "q8_0_128k_total_gib": 343.42, "fp16_8k_total_gib": 624.48, "fp16_32k_total_gib": 626.69, "fp16_128k_total_gib": 635.56, "kv_cache_8k_gib": 0.67, "kv_cache_32k_gib": 2.69, "kv_cache_128k_gib": 10.75, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "This model shares and compresses its KV cache across layers, which the calculator does not model, so the KV cache figure is an upper bound." }, { "model": "Qwen3.6 35B-A3B (MoE)", "hf_id": "Qwen/Qwen3.6-35B-A3B", "hf_url": "https://huggingface.co/Qwen/Qwen3.6-35B-A3B", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3.6-35b-a3b/", "source_checked": "2026-09-26", "total_params": 35951822704, "active_params": 3000000000, "experts": 256, "layers": 40, "max_context": 262144, "kv_bytes_per_token_fp16": 20480, "attention_layout": "10 of its 40 layers use full attention and 30 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 20.26, "q8_0_weights_gib": 35.58, "fp16_weights_gib": 66.97, "q4_k_m_8k_total_gib": 22.95, "q4_k_m_32k_total_gib": 23.47, "q4_k_m_128k_total_gib": 25.53, "q8_0_8k_total_gib": 39.8, "q8_0_32k_total_gib": 40.32, "q8_0_128k_total_gib": 42.38, "fp16_8k_total_gib": 74.33, "fp16_32k_total_gib": 74.85, "fp16_128k_total_gib": 76.91, "kv_cache_8k_gib": 0.16, "kv_cache_32k_gib": 0.63, "kv_cache_128k_gib": 2.5, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Qwen3.6 27B", "hf_id": "Qwen/Qwen3.6-27B", "hf_url": "https://huggingface.co/Qwen/Qwen3.6-27B", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3.6-27b/", "source_checked": "2026-09-26", "total_params": 27781427952, "active_params": null, "experts": null, "layers": 64, "max_context": 262144, "kv_bytes_per_token_fp16": 65536, "attention_layout": "16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 15.65, "q8_0_weights_gib": 27.49, "fp16_weights_gib": 51.75, "q4_k_m_8k_total_gib": 18.27, "q4_k_m_32k_total_gib": 19.92, "q4_k_m_128k_total_gib": 26.52, "q8_0_8k_total_gib": 31.29, "q8_0_32k_total_gib": 32.94, "q8_0_128k_total_gib": 39.54, "fp16_8k_total_gib": 57.97, "fp16_32k_total_gib": 59.62, "fp16_128k_total_gib": 66.22, "kv_cache_8k_gib": 0.5, "kv_cache_32k_gib": 2, "kv_cache_128k_gib": 8, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Qwen3.8 27B", "hf_id": "Qwen/Qwen3.8-27B", "hf_url": "https://huggingface.co/Qwen/Qwen3.8-27B", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3.8-27b/", "source_checked": "2026-09-26", "total_params": 27781427952, "active_params": null, "experts": null, "layers": 64, "max_context": 262144, "kv_bytes_per_token_fp16": 65536, "attention_layout": "16 of its 64 layers use full attention and 48 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 15.65, "q8_0_weights_gib": 27.49, "fp16_weights_gib": 51.75, "q4_k_m_8k_total_gib": 18.27, "q4_k_m_32k_total_gib": 19.92, "q4_k_m_128k_total_gib": 26.52, "q8_0_8k_total_gib": 31.29, "q8_0_32k_total_gib": 32.94, "q8_0_128k_total_gib": 39.54, "fp16_8k_total_gib": 57.97, "fp16_32k_total_gib": 59.62, "fp16_128k_total_gib": 66.22, "kv_cache_8k_gib": 0.5, "kv_cache_32k_gib": 2, "kv_cache_128k_gib": 8, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Qwen3.8 Flash Next (180B MoE)", "hf_id": "Qwen/Qwen3.8-Flash-Next", "hf_url": "https://huggingface.co/Qwen/Qwen3.8-Flash-Next", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3.8-flash-next/", "source_checked": "2026-09-26", "total_params": 179999981459, "active_params": 6000000000, "experts": 512, "layers": 48, "max_context": 262144, "kv_bytes_per_token_fp16": 24576, "attention_layout": "12 of its 48 layers use full attention and 36 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 101.42, "q8_0_weights_gib": 178.12, "fp16_weights_gib": 335.28, "q4_k_m_8k_total_gib": 112.27, "q4_k_m_32k_total_gib": 112.89, "q4_k_m_128k_total_gib": 115.36, "q8_0_8k_total_gib": 196.63, "q8_0_32k_total_gib": 197.25, "q8_0_128k_total_gib": 199.73, "fp16_8k_total_gib": 369.51, "fp16_32k_total_gib": 370.13, "fp16_128k_total_gib": 372.6, "kv_cache_8k_gib": 0.19, "kv_cache_32k_gib": 0.75, "kv_cache_128k_gib": 3, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Qwen3.8 2.4T-A95B (MoE)", "hf_id": "Qwen/Qwen3.8-2.4T-A95B", "hf_url": "https://huggingface.co/Qwen/Qwen3.8-2.4T-A95B", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3.8-2.4t-a95b/", "source_checked": "2026-09-26", "total_params": 2446182725504, "active_params": 95000000000, "experts": 512, "layers": 92, "max_context": 262144, "kv_bytes_per_token_fp16": 94208, "attention_layout": "23 of its 92 layers use full attention and 69 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 1378.3, "q8_0_weights_gib": 2420.57, "fp16_weights_gib": 4556.37, "q4_k_m_8k_total_gib": 1517.42, "q4_k_m_32k_total_gib": 1519.79, "q4_k_m_128k_total_gib": 1529.28, "q8_0_8k_total_gib": 2663.92, "q8_0_32k_total_gib": 2666.29, "q8_0_128k_total_gib": 2675.78, "fp16_8k_total_gib": 5013.3, "fp16_32k_total_gib": 5015.67, "fp16_128k_total_gib": 5025.16, "kv_cache_8k_gib": 0.72, "kv_cache_32k_gib": 2.88, "kv_cache_128k_gib": 11.5, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Qwen3.5 9B", "hf_id": "Qwen/Qwen3.5-9B", "hf_url": "https://huggingface.co/Qwen/Qwen3.5-9B", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3.5-9b/", "source_checked": "2026-09-26", "total_params": 9653104368, "active_params": null, "experts": null, "layers": 32, "max_context": 262144, "kv_bytes_per_token_fp16": 32768, "attention_layout": "8 of its 32 layers use full attention and 24 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 5.44, "q8_0_weights_gib": 9.55, "fp16_weights_gib": 17.98, "q4_k_m_8k_total_gib": 6.76, "q4_k_m_32k_total_gib": 7.58, "q4_k_m_128k_total_gib": 10.88, "q8_0_8k_total_gib": 11.28, "q8_0_32k_total_gib": 12.11, "q8_0_128k_total_gib": 15.41, "fp16_8k_total_gib": 20.55, "fp16_32k_total_gib": 21.38, "fp16_128k_total_gib": 24.68, "kv_cache_8k_gib": 0.25, "kv_cache_32k_gib": 1, "kv_cache_128k_gib": 4, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Qwen3.5 122B-A10B (MoE)", "hf_id": "Qwen/Qwen3.5-122B-A10B", "hf_url": "https://huggingface.co/Qwen/Qwen3.5-122B-A10B", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3.5-122b-a10b/", "source_checked": "2026-09-26", "total_params": 125086497008, "active_params": 10000000000, "experts": 256, "layers": 48, "max_context": 262144, "kv_bytes_per_token_fp16": 24576, "attention_layout": "12 of its 48 layers use full attention and 36 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 70.48, "q8_0_weights_gib": 123.78, "fp16_weights_gib": 232.99, "q4_k_m_8k_total_gib": 78.23, "q4_k_m_32k_total_gib": 78.85, "q4_k_m_128k_total_gib": 81.33, "q8_0_8k_total_gib": 136.86, "q8_0_32k_total_gib": 137.48, "q8_0_128k_total_gib": 139.95, "fp16_8k_total_gib": 257, "fp16_32k_total_gib": 257.62, "fp16_128k_total_gib": 260.09, "kv_cache_8k_gib": 0.19, "kv_cache_32k_gib": 0.75, "kv_cache_128k_gib": 3, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Qwen3-Coder-Next (80B MoE)", "hf_id": "Qwen/Qwen3-Coder-Next", "hf_url": "https://huggingface.co/Qwen/Qwen3-Coder-Next", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3-coder-next/", "source_checked": "2026-09-26", "total_params": 79674391296, "active_params": 3000000000, "experts": 512, "layers": 48, "max_context": 262144, "kv_bytes_per_token_fp16": 24576, "attention_layout": "12 of its 48 layers use full attention and 36 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 44.89, "q8_0_weights_gib": 78.84, "fp16_weights_gib": 148.41, "q4_k_m_8k_total_gib": 50.09, "q4_k_m_32k_total_gib": 50.71, "q4_k_m_128k_total_gib": 53.18, "q8_0_8k_total_gib": 87.43, "q8_0_32k_total_gib": 88.05, "q8_0_128k_total_gib": 90.52, "fp16_8k_total_gib": 163.95, "fp16_32k_total_gib": 164.57, "fp16_128k_total_gib": 167.05, "kv_cache_8k_gib": 0.19, "kv_cache_32k_gib": 0.75, "kv_cache_128k_gib": 3, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "DeepSeek V4.1 Flash", "hf_id": "deepseek-ai/DeepSeek-V4.1-Flash", "hf_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash", "model_url": "https://modelvram.com/llm-vram-calculator/deepseek-v4.1-flash/", "source_checked": "2026-09-26", "total_params": 763205315794, "active_params": 16000000000, "experts": 384, "layers": 40, "max_context": 1048576, "kv_bytes_per_token_fp16": 81920, "attention_layout": "All 40 layers use full attention", "q4_k_m_weights_gib": 430.03, "q8_0_weights_gib": 755.21, "fp16_weights_gib": 1421.58, "q4_k_m_8k_total_gib": 474.22, "q4_k_m_32k_total_gib": 476.28, "q4_k_m_128k_total_gib": 484.53, "q8_0_8k_total_gib": 831.92, "q8_0_32k_total_gib": 833.99, "q8_0_128k_total_gib": 842.24, "fp16_8k_total_gib": 1564.93, "fp16_32k_total_gib": 1566.99, "fp16_128k_total_gib": 1575.24, "kv_cache_8k_gib": 0.63, "kv_cache_32k_gib": 2.5, "kv_cache_128k_gib": 10, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "This model shares and compresses its KV cache across layers, which the calculator does not model, so the KV cache figure is an upper bound." }, { "model": "DeepSeek V4 Flash", "hf_id": "deepseek-ai/DeepSeek-V4-Flash", "hf_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash", "model_url": "https://modelvram.com/llm-vram-calculator/deepseek-v4-flash/", "source_checked": "2026-09-26", "total_params": 290944616402, "active_params": 13000000000, "experts": 256, "layers": 43, "max_context": 1048576, "kv_bytes_per_token_fp16": 88064, "attention_layout": "All 43 layers use full attention", "q4_k_m_weights_gib": 163.93, "q8_0_weights_gib": 287.9, "fp16_weights_gib": 541.93, "q4_k_m_8k_total_gib": 181.57, "q4_k_m_32k_total_gib": 183.78, "q4_k_m_128k_total_gib": 192.65, "q8_0_8k_total_gib": 317.93, "q8_0_32k_total_gib": 320.14, "q8_0_128k_total_gib": 329.01, "fp16_8k_total_gib": 597.36, "fp16_32k_total_gib": 599.58, "fp16_128k_total_gib": 608.44, "kv_cache_8k_gib": 0.67, "kv_cache_32k_gib": 2.69, "kv_cache_128k_gib": 10.75, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "This model shares and compresses its KV cache across layers, which the calculator does not model, so the KV cache figure is an upper bound." }, { "model": "DeepSeek V4 Pro", "hf_id": "deepseek-ai/DeepSeek-V4-Pro", "hf_url": "https://huggingface.co/deepseek-ai/DeepSeek-V4-Pro", "model_url": "https://modelvram.com/llm-vram-calculator/deepseek-v4-pro/", "source_checked": "2026-09-26", "total_params": 1598839674782, "active_params": 49000000000, "experts": 384, "layers": 61, "max_context": 1048576, "kv_bytes_per_token_fp16": 124928, "attention_layout": "All 61 layers use full attention", "q4_k_m_weights_gib": 900.87, "q8_0_weights_gib": 1582.1, "fp16_weights_gib": 2978.07, "q4_k_m_8k_total_gib": 992.5, "q4_k_m_32k_total_gib": 995.65, "q4_k_m_128k_total_gib": 1008.23, "q8_0_8k_total_gib": 1741.86, "q8_0_32k_total_gib": 1745, "q8_0_128k_total_gib": 1757.59, "fp16_8k_total_gib": 3277.43, "fp16_32k_total_gib": 3280.57, "fp16_128k_total_gib": 3293.15, "kv_cache_8k_gib": 0.95, "kv_cache_32k_gib": 3.81, "kv_cache_128k_gib": 15.25, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "This model shares and compresses its KV cache across layers, which the calculator does not model, so the KV cache figure is an upper bound." }, { "model": "DeepSeek V3.2", "hf_id": "deepseek-ai/DeepSeek-V3.2", "hf_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3.2", "model_url": "https://modelvram.com/llm-vram-calculator/deepseek-v3.2/", "source_checked": "2026-09-26", "total_params": 685355329792, "active_params": 37000000000, "experts": 256, "layers": 61, "max_context": 163840, "kv_bytes_per_token_fp16": 78080, "attention_layout": "All 61 layers use multi-head latent attention", "q4_k_m_weights_gib": 386.16, "q8_0_weights_gib": 678.18, "fp16_weights_gib": 1276.57, "q4_k_m_8k_total_gib": 425.94, "q4_k_m_32k_total_gib": 427.9, "q4_k_m_128k_total_gib": 435.76, "q8_0_8k_total_gib": 747.15, "q8_0_32k_total_gib": 749.12, "q8_0_128k_total_gib": 756.98, "fp16_8k_total_gib": 1405.39, "fp16_32k_total_gib": 1407.35, "fp16_128k_total_gib": 1415.22, "kv_cache_8k_gib": 0.6, "kv_cache_32k_gib": 2.38, "kv_cache_128k_gib": 9.53, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "GLM-5.3", "hf_id": "zai-org/GLM-5.3", "hf_url": "https://huggingface.co/zai-org/GLM-5.3", "model_url": "https://modelvram.com/llm-vram-calculator/glm-5.3/", "source_checked": "2026-09-26", "total_params": 753329940480, "active_params": 40000000000, "experts": 256, "layers": 78, "max_context": 1048576, "kv_bytes_per_token_fp16": 92544, "attention_layout": "All 78 layers use multi-head latent attention", "q4_k_m_weights_gib": 424.46, "q8_0_weights_gib": 745.44, "fp16_weights_gib": 1403.19, "q4_k_m_8k_total_gib": 468.19, "q4_k_m_32k_total_gib": 470.52, "q4_k_m_128k_total_gib": 479.84, "q8_0_8k_total_gib": 821.26, "q8_0_32k_total_gib": 823.59, "q8_0_128k_total_gib": 832.91, "fp16_8k_total_gib": 1544.78, "fp16_32k_total_gib": 1547.11, "fp16_128k_total_gib": 1556.43, "kv_cache_8k_gib": 0.71, "kv_cache_32k_gib": 2.82, "kv_cache_128k_gib": 11.3, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "GLM-5.3 Flash", "hf_id": "zai-org/GLM-5.3-Flash", "hf_url": "https://huggingface.co/zai-org/GLM-5.3-Flash", "model_url": "https://modelvram.com/llm-vram-calculator/glm-5.3-flash/", "source_checked": "2026-09-26", "total_params": 321323031390, "active_params": 18000000000, "experts": 288, "layers": 45, "max_context": 1048576, "kv_bytes_per_token_fp16": 12672, "attention_layout": "11 of its 45 layers use multi-head latent attention and 34 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 181.05, "q8_0_weights_gib": 317.96, "fp16_weights_gib": 598.51, "q4_k_m_8k_total_gib": 199.76, "q4_k_m_32k_total_gib": 200.08, "q4_k_m_128k_total_gib": 201.36, "q8_0_8k_total_gib": 350.36, "q8_0_32k_total_gib": 350.68, "q8_0_128k_total_gib": 351.96, "fp16_8k_total_gib": 658.97, "fp16_32k_total_gib": 659.29, "fp16_128k_total_gib": 660.56, "kv_cache_8k_gib": 0.1, "kv_cache_32k_gib": 0.39, "kv_cache_128k_gib": 1.55, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "GLM-5.2", "hf_id": "zai-org/GLM-5.2", "hf_url": "https://huggingface.co/zai-org/GLM-5.2", "model_url": "https://modelvram.com/llm-vram-calculator/glm-5.2/", "source_checked": "2026-09-26", "total_params": 753329940480, "active_params": 40000000000, "experts": 256, "layers": 78, "max_context": 1048576, "kv_bytes_per_token_fp16": 92544, "attention_layout": "All 78 layers use multi-head latent attention", "q4_k_m_weights_gib": 424.46, "q8_0_weights_gib": 745.44, "fp16_weights_gib": 1403.19, "q4_k_m_8k_total_gib": 468.19, "q4_k_m_32k_total_gib": 470.52, "q4_k_m_128k_total_gib": 479.84, "q8_0_8k_total_gib": 821.26, "q8_0_32k_total_gib": 823.59, "q8_0_128k_total_gib": 832.91, "fp16_8k_total_gib": 1544.78, "fp16_32k_total_gib": 1547.11, "fp16_128k_total_gib": 1556.43, "kv_cache_8k_gib": 0.71, "kv_cache_32k_gib": 2.82, "kv_cache_128k_gib": 11.3, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "GLM-4.7 Flash", "hf_id": "zai-org/GLM-4.7-Flash", "hf_url": "https://huggingface.co/zai-org/GLM-4.7-Flash", "model_url": "https://modelvram.com/llm-vram-calculator/glm-4.7-flash/", "source_checked": "2026-09-26", "total_params": 31221488576, "active_params": 3000000000, "experts": 64, "layers": 47, "max_context": 202752, "kv_bytes_per_token_fp16": 54144, "attention_layout": "All 47 layers use multi-head latent attention", "q4_k_m_weights_gib": 17.59, "q8_0_weights_gib": 30.89, "fp16_weights_gib": 58.15, "q4_k_m_8k_total_gib": 20.31, "q4_k_m_32k_total_gib": 21.67, "q4_k_m_128k_total_gib": 27.12, "q8_0_8k_total_gib": 34.94, "q8_0_32k_total_gib": 36.3, "q8_0_128k_total_gib": 41.75, "fp16_8k_total_gib": 64.92, "fp16_32k_total_gib": 66.29, "fp16_128k_total_gib": 71.74, "kv_cache_8k_gib": 0.41, "kv_cache_32k_gib": 1.65, "kv_cache_128k_gib": 6.61, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Gemma 4 31B", "hf_id": "google/gemma-4-31B-it", "hf_url": "https://huggingface.co/google/gemma-4-31B-it", "model_url": "https://modelvram.com/llm-vram-calculator/gemma-4-31b-it/", "source_checked": "2026-09-26", "total_params": 31273088876, "active_params": null, "experts": null, "layers": 60, "max_context": 262144, "kv_bytes_per_token_fp16": 81920, "attention_layout": "10 of its 60 layers use full attention and 50 keep a sliding window of 1,024 tokens", "q4_k_m_weights_gib": 17.62, "q8_0_weights_gib": 30.95, "fp16_weights_gib": 58.25, "q4_k_m_8k_total_gib": 21.86, "q4_k_m_32k_total_gib": 23.92, "q4_k_m_128k_total_gib": 32.17, "q8_0_8k_total_gib": 36.52, "q8_0_32k_total_gib": 38.58, "q8_0_128k_total_gib": 46.83, "fp16_8k_total_gib": 66.55, "fp16_32k_total_gib": 68.61, "fp16_128k_total_gib": 76.86, "kv_cache_8k_gib": 1.8, "kv_cache_32k_gib": 3.67, "kv_cache_128k_gib": 11.17, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Gemma 4 26B-A4B (MoE)", "hf_id": "google/gemma-4-26B-A4B-it", "hf_url": "https://huggingface.co/google/gemma-4-26B-A4B-it", "model_url": "https://modelvram.com/llm-vram-calculator/gemma-4-26b-a4b-it/", "source_checked": "2026-09-26", "total_params": 25805936206, "active_params": 4000000000, "experts": 128, "layers": 30, "max_context": 262144, "kv_bytes_per_token_fp16": 20480, "attention_layout": "5 of its 30 layers use full attention and 25 keep a sliding window of 1,024 tokens", "q4_k_m_weights_gib": 14.54, "q8_0_weights_gib": 25.54, "fp16_weights_gib": 48.07, "q4_k_m_8k_total_gib": 16.99, "q4_k_m_32k_total_gib": 17.5, "q4_k_m_128k_total_gib": 19.57, "q8_0_8k_total_gib": 29.08, "q8_0_32k_total_gib": 29.6, "q8_0_128k_total_gib": 31.66, "fp16_8k_total_gib": 53.87, "fp16_32k_total_gib": 54.38, "fp16_128k_total_gib": 56.45, "kv_cache_8k_gib": 0.45, "kv_cache_32k_gib": 0.92, "kv_cache_128k_gib": 2.79, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Gemma 4 12B", "hf_id": "google/gemma-4-12B-it", "hf_url": "https://huggingface.co/google/gemma-4-12B-it", "model_url": "https://modelvram.com/llm-vram-calculator/gemma-4-12b-it/", "source_checked": "2026-09-26", "total_params": 11959730224, "active_params": null, "experts": null, "layers": 48, "max_context": 262144, "kv_bytes_per_token_fp16": 16384, "attention_layout": "8 of its 48 layers use full attention and 40 keep a sliding window of 1,024 tokens", "q4_k_m_weights_gib": 6.74, "q8_0_weights_gib": 11.83, "fp16_weights_gib": 22.28, "q4_k_m_8k_total_gib": 8.57, "q4_k_m_32k_total_gib": 8.98, "q4_k_m_128k_total_gib": 10.63, "q8_0_8k_total_gib": 14.17, "q8_0_32k_total_gib": 14.58, "q8_0_128k_total_gib": 16.23, "fp16_8k_total_gib": 25.66, "fp16_32k_total_gib": 26.07, "fp16_128k_total_gib": 27.72, "kv_cache_8k_gib": 0.59, "kv_cache_32k_gib": 0.97, "kv_cache_128k_gib": 2.47, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Gemma 4 E4B", "hf_id": "google/gemma-4-E4B-it", "hf_url": "https://huggingface.co/google/gemma-4-E4B-it", "model_url": "https://modelvram.com/llm-vram-calculator/gemma-4-e4b-it/", "source_checked": "2026-09-26", "total_params": 7996156490, "active_params": null, "experts": null, "layers": 42, "max_context": 131072, "kv_bytes_per_token_fp16": 16384, "attention_layout": "4 of its 42 layers use full attention, 20 keep a sliding window of 512 tokens and 18 reuse the cache of earlier layers", "q4_k_m_weights_gib": 4.51, "q8_0_weights_gib": 7.91, "fp16_weights_gib": 14.89, "q4_k_m_8k_total_gib": 5.64, "q4_k_m_32k_total_gib": 6.05, "q4_k_m_128k_total_gib": 7.7, "q8_0_8k_total_gib": 9.38, "q8_0_32k_total_gib": 9.8, "q8_0_128k_total_gib": 11.45, "fp16_8k_total_gib": 17.06, "fp16_32k_total_gib": 17.48, "fp16_128k_total_gib": 19.13, "kv_cache_8k_gib": 0.16, "kv_cache_32k_gib": 0.54, "kv_cache_128k_gib": 2.04, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Kimi K3", "hf_id": "moonshotai/Kimi-K3", "hf_url": "https://huggingface.co/moonshotai/Kimi-K3", "model_url": "https://modelvram.com/llm-vram-calculator/kimi-k3/", "source_checked": "2026-09-26", "total_params": 2779931837184, "active_params": 104000000000, "experts": 896, "layers": 93, "max_context": 1048576, "kv_bytes_per_token_fp16": 27648, "attention_layout": "24 of its 93 layers use multi-head latent attention and 69 are linear-attention layers with no growing cache", "q4_k_m_weights_gib": 1566.35, "q8_0_weights_gib": 2750.83, "fp16_weights_gib": 5178.03, "q4_k_m_8k_total_gib": 1723.72, "q4_k_m_32k_total_gib": 1724.42, "q4_k_m_128k_total_gib": 1727.2, "q8_0_8k_total_gib": 3026.64, "q8_0_32k_total_gib": 3027.34, "q8_0_128k_total_gib": 3030.12, "fp16_8k_total_gib": 5696.56, "fp16_32k_total_gib": 5697.26, "fp16_128k_total_gib": 5700.04, "kv_cache_8k_gib": 0.21, "kv_cache_32k_gib": 0.84, "kv_cache_128k_gib": 3.38, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "MiniMax M3", "hf_id": "MiniMaxAI/MiniMax-M3", "hf_url": "https://huggingface.co/MiniMaxAI/MiniMax-M3", "model_url": "https://modelvram.com/llm-vram-calculator/minimax-m3/", "source_checked": "2026-09-26", "total_params": 427040140160, "active_params": 23000000000, "experts": 128, "layers": 60, "max_context": 1048576, "kv_bytes_per_token_fp16": 122880, "attention_layout": "All 60 layers use full attention", "q4_k_m_weights_gib": 240.62, "q8_0_weights_gib": 422.57, "fp16_weights_gib": 795.42, "q4_k_m_8k_total_gib": 266.21, "q4_k_m_32k_total_gib": 269.3, "q4_k_m_128k_total_gib": 281.68, "q8_0_8k_total_gib": 466.36, "q8_0_32k_total_gib": 469.45, "q8_0_128k_total_gib": 481.83, "fp16_8k_total_gib": 876.5, "fp16_32k_total_gib": 879.59, "fp16_128k_total_gib": 891.97, "kv_cache_8k_gib": 0.94, "kv_cache_32k_gib": 3.75, "kv_cache_128k_gib": 15, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "MiniMax M2.7", "hf_id": "MiniMaxAI/MiniMax-M2.7", "hf_url": "https://huggingface.co/MiniMaxAI/MiniMax-M2.7", "model_url": "https://modelvram.com/llm-vram-calculator/minimax-m2.7/", "source_checked": "2026-09-26", "total_params": 228689764864, "active_params": 10000000000, "experts": 256, "layers": 62, "max_context": 204800, "kv_bytes_per_token_fp16": 253952, "attention_layout": "All 62 layers use full attention", "q4_k_m_weights_gib": 128.86, "q8_0_weights_gib": 226.3, "fp16_weights_gib": 425.97, "q4_k_m_8k_total_gib": 144.37, "q4_k_m_32k_total_gib": 150.77, "q4_k_m_128k_total_gib": 176.34, "q8_0_8k_total_gib": 251.56, "q8_0_32k_total_gib": 257.95, "q8_0_128k_total_gib": 283.52, "fp16_8k_total_gib": 471.2, "fp16_32k_total_gib": 477.59, "fp16_128k_total_gib": 503.16, "kv_cache_8k_gib": 1.94, "kv_cache_32k_gib": 7.75, "kv_cache_128k_gib": 31, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "MiMo V2.6 Flash", "hf_id": "XiaomiMiMo/MiMo-V2.6-Flash-RL", "hf_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Flash-RL", "model_url": "https://modelvram.com/llm-vram-calculator/mimo-v2.6-flash-rl/", "source_checked": "2026-09-26", "total_params": 310756322688, "active_params": 15000000000, "experts": 256, "layers": 48, "max_context": 1048576, "kv_bytes_per_token_fp16": 23040, "attention_layout": "9 of its 48 layers use full attention and 39 keep a sliding window of 128 tokens", "q4_k_m_weights_gib": 175.1, "q8_0_weights_gib": 307.5, "fp16_weights_gib": 578.83, "q4_k_m_8k_total_gib": 193.46, "q4_k_m_32k_total_gib": 194.04, "q4_k_m_128k_total_gib": 196.36, "q8_0_8k_total_gib": 339.1, "q8_0_32k_total_gib": 339.68, "q8_0_128k_total_gib": 342, "fp16_8k_total_gib": 637.56, "fp16_32k_total_gib": 638.14, "fp16_128k_total_gib": 640.46, "kv_cache_8k_gib": 0.32, "kv_cache_32k_gib": 0.85, "kv_cache_128k_gib": 2.96, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "MiMo V2.6 Pro", "hf_id": "XiaomiMiMo/MiMo-V2.6-Pro-RL", "hf_url": "https://huggingface.co/XiaomiMiMo/MiMo-V2.6-Pro-RL", "model_url": "https://modelvram.com/llm-vram-calculator/mimo-v2.6-pro-rl/", "source_checked": "2026-09-26", "total_params": 1024216603392, "active_params": 42000000000, "experts": 384, "layers": 70, "max_context": 1048576, "kv_bytes_per_token_fp16": 51200, "attention_layout": "10 of its 70 layers use full attention and 60 keep a sliding window of 128 tokens", "q4_k_m_weights_gib": 577.1, "q8_0_weights_gib": 1013.49, "fp16_weights_gib": 1907.75, "q4_k_m_8k_total_gib": 635.98, "q4_k_m_32k_total_gib": 637.26, "q4_k_m_128k_total_gib": 642.42, "q8_0_8k_total_gib": 1116.01, "q8_0_32k_total_gib": 1117.3, "q8_0_128k_total_gib": 1122.46, "fp16_8k_total_gib": 2099.7, "fp16_32k_total_gib": 2100.99, "fp16_128k_total_gib": 2106.14, "kv_cache_8k_gib": 0.61, "kv_cache_32k_gib": 1.78, "kv_cache_128k_gib": 6.47, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Mistral Medium 3.5 128B", "hf_id": "mistralai/Mistral-Medium-3.5-128B", "hf_url": "https://huggingface.co/mistralai/Mistral-Medium-3.5-128B", "model_url": "https://modelvram.com/llm-vram-calculator/mistral-medium-3.5-128b/", "source_checked": "2026-09-26", "total_params": 127704210176, "active_params": null, "experts": null, "layers": 88, "max_context": 262144, "kv_bytes_per_token_fp16": 360448, "attention_layout": "All 88 layers use full attention", "q4_k_m_weights_gib": 71.95, "q8_0_weights_gib": 126.37, "fp16_weights_gib": 237.87, "q4_k_m_8k_total_gib": 82.68, "q4_k_m_32k_total_gib": 91.75, "q4_k_m_128k_total_gib": 128.05, "q8_0_8k_total_gib": 142.53, "q8_0_32k_total_gib": 151.6, "q8_0_128k_total_gib": 187.9, "fp16_8k_total_gib": 265.18, "fp16_32k_total_gib": 274.25, "fp16_128k_total_gib": 310.55, "kv_cache_8k_gib": 2.75, "kv_cache_32k_gib": 11, "kv_cache_128k_gib": 44, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Nemotron 3 Nano 4B", "hf_id": "nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16", "hf_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16", "model_url": "https://modelvram.com/llm-vram-calculator/nemotron-3-nano-4b/", "source_checked": "2026-09-26", "total_params": 3973556832, "active_params": null, "experts": null, "layers": 42, "max_context": 262144, "kv_bytes_per_token_fp16": 16384, "attention_layout": "4 of its 42 layers use full attention and 38 are Mamba or feed-forward layers with no growing cache", "q4_k_m_weights_gib": 2.24, "q8_0_weights_gib": 3.93, "fp16_weights_gib": 7.4, "q4_k_m_8k_total_gib": 3.1, "q4_k_m_32k_total_gib": 3.51, "q4_k_m_128k_total_gib": 5.16, "q8_0_8k_total_gib": 4.96, "q8_0_32k_total_gib": 5.38, "q8_0_128k_total_gib": 7.03, "fp16_8k_total_gib": 8.78, "fp16_32k_total_gib": 9.19, "fp16_128k_total_gib": 10.84, "kv_cache_8k_gib": 0.13, "kv_cache_32k_gib": 0.5, "kv_cache_128k_gib": 2, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": true, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Nemotron 3 Nano 30B-A3B (MoE)", "hf_id": "nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "hf_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "model_url": "https://modelvram.com/llm-vram-calculator/nemotron-3-nano-30b-a3b/", "source_checked": "2026-09-26", "total_params": 31577937344, "active_params": 3500000000, "experts": 128, "layers": 52, "max_context": 262144, "kv_bytes_per_token_fp16": 6144, "attention_layout": "6 of its 52 layers use full attention and 46 are Mamba or feed-forward layers with no growing cache", "q4_k_m_weights_gib": 17.79, "q8_0_weights_gib": 31.25, "fp16_weights_gib": 58.82, "q4_k_m_8k_total_gib": 20.12, "q4_k_m_32k_total_gib": 20.28, "q4_k_m_128k_total_gib": 20.9, "q8_0_8k_total_gib": 34.92, "q8_0_32k_total_gib": 35.08, "q8_0_128k_total_gib": 35.7, "fp16_8k_total_gib": 65.25, "fp16_32k_total_gib": 65.41, "fp16_128k_total_gib": 66.03, "kv_cache_8k_gib": 0.05, "kv_cache_32k_gib": 0.19, "kv_cache_128k_gib": 0.75, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Nemotron 3 Super 120B-A12B (MoE)", "hf_id": "nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "hf_url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "model_url": "https://modelvram.com/llm-vram-calculator/nemotron-3-super-120b-a12b/", "source_checked": "2026-09-26", "total_params": 123611012096, "active_params": 12000000000, "experts": 512, "layers": 88, "max_context": 262144, "kv_bytes_per_token_fp16": 8192, "attention_layout": "8 of its 88 layers use full attention and 80 are Mamba or feed-forward layers with no growing cache", "q4_k_m_weights_gib": 69.65, "q8_0_weights_gib": 122.32, "fp16_weights_gib": 230.24, "q4_k_m_8k_total_gib": 77.18, "q4_k_m_32k_total_gib": 77.39, "q4_k_m_128k_total_gib": 78.21, "q8_0_8k_total_gib": 135.12, "q8_0_32k_total_gib": 135.32, "q8_0_128k_total_gib": 136.15, "fp16_8k_total_gib": 253.84, "fp16_32k_total_gib": 254.04, "fp16_128k_total_gib": 254.87, "kv_cache_8k_gib": 0.06, "kv_cache_32k_gib": 0.25, "kv_cache_128k_gib": 1, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Xing 4.0 29B-A4B (MoE)", "hf_id": "XingChen-AGI/Xing4.0-29B-A4B", "hf_url": "https://huggingface.co/XingChen-AGI/Xing4.0-29B-A4B", "model_url": "https://modelvram.com/llm-vram-calculator/xing4.0-29b-a4b/", "source_checked": "2026-09-26", "total_params": 31215031088, "active_params": 4000000000, "experts": 64, "layers": 40, "max_context": 262144, "kv_bytes_per_token_fp16": 46080, "attention_layout": "All 40 layers use multi-head latent attention", "q4_k_m_weights_gib": 17.59, "q8_0_weights_gib": 30.89, "fp16_weights_gib": 58.14, "q4_k_m_8k_total_gib": 20.23, "q4_k_m_32k_total_gib": 21.39, "q4_k_m_128k_total_gib": 26.03, "q8_0_8k_total_gib": 34.86, "q8_0_32k_total_gib": 36.02, "q8_0_128k_total_gib": 40.66, "fp16_8k_total_gib": 64.84, "fp16_32k_total_gib": 66, "fp16_128k_total_gib": 70.64, "kv_cache_8k_gib": 0.35, "kv_cache_32k_gib": 1.41, "kv_cache_128k_gib": 5.63, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Muse Glimmer 30B", "hf_id": "meta-models/Muse-Glimmer-30B", "hf_url": "https://huggingface.co/meta-models/Muse-Glimmer-30B", "model_url": "https://modelvram.com/llm-vram-calculator/muse-glimmer-30b/", "source_checked": "2026-09-26", "total_params": 29776626688, "active_params": null, "experts": null, "layers": 52, "max_context": 131072, "kv_bytes_per_token_fp16": 13312, "attention_layout": "13 of its 52 layers use full attention and 39 keep a sliding window of 2,048 tokens", "q4_k_m_weights_gib": 16.78, "q8_0_weights_gib": 29.46, "fp16_weights_gib": 55.46, "q4_k_m_8k_total_gib": 19.17, "q4_k_m_32k_total_gib": 19.51, "q4_k_m_128k_total_gib": 20.85, "q8_0_8k_total_gib": 33.13, "q8_0_32k_total_gib": 33.46, "q8_0_128k_total_gib": 34.8, "fp16_8k_total_gib": 61.73, "fp16_32k_total_gib": 62.06, "fp16_128k_total_gib": 63.4, "kv_cache_8k_gib": 0.2, "kv_cache_32k_gib": 0.5, "kv_cache_128k_gib": 1.72, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "MiniCPM5 2B", "hf_id": "openbmb/MiniCPM5-2B", "hf_url": "https://huggingface.co/openbmb/MiniCPM5-2B", "model_url": "https://modelvram.com/llm-vram-calculator/minicpm5-2b/", "source_checked": "2026-09-26", "total_params": 2516756480, "active_params": null, "experts": null, "layers": 42, "max_context": 131072, "kv_bytes_per_token_fp16": 43008, "attention_layout": "All 42 layers use full attention", "q4_k_m_weights_gib": 1.42, "q8_0_weights_gib": 2.49, "fp16_weights_gib": 4.69, "q4_k_m_8k_total_gib": 2.42, "q4_k_m_32k_total_gib": 3.5, "q4_k_m_128k_total_gib": 7.83, "q8_0_8k_total_gib": 3.6, "q8_0_32k_total_gib": 4.68, "q8_0_128k_total_gib": 9.01, "fp16_8k_total_gib": 6.02, "fp16_32k_total_gib": 7.1, "fp16_128k_total_gib": 11.43, "kv_cache_8k_gib": 0.33, "kv_cache_32k_gib": 1.31, "kv_cache_128k_gib": 5.25, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": true, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Llama 3.1 8B", "hf_id": "meta-llama/Llama-3.1-8B-Instruct", "hf_url": "https://huggingface.co/meta-llama/Llama-3.1-8B-Instruct", "model_url": "https://modelvram.com/llm-vram-calculator/llama-3.1-8b-instruct/", "source_checked": "2026-09-26", "total_params": 8030261248, "active_params": null, "experts": null, "layers": 32, "max_context": 131072, "kv_bytes_per_token_fp16": 131072, "attention_layout": "All 32 layers use full attention", "q4_k_m_weights_gib": 4.52, "q8_0_weights_gib": 7.95, "fp16_weights_gib": 14.96, "q4_k_m_8k_total_gib": 6.58, "q4_k_m_32k_total_gib": 9.88, "q4_k_m_128k_total_gib": 23.08, "q8_0_8k_total_gib": 10.34, "q8_0_32k_total_gib": 13.64, "q8_0_128k_total_gib": 26.84, "fp16_8k_total_gib": 18.05, "fp16_32k_total_gib": 21.35, "fp16_128k_total_gib": 34.55, "kv_cache_8k_gib": 1, "kv_cache_32k_gib": 4, "kv_cache_128k_gib": 16, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Llama 3.1 70B", "hf_id": "meta-llama/Llama-3.1-70B-Instruct", "hf_url": "https://huggingface.co/meta-llama/Llama-3.1-70B-Instruct", "model_url": "https://modelvram.com/llm-vram-calculator/llama-3.1-70b-instruct/", "source_checked": "2026-09-26", "total_params": 70553706496, "active_params": null, "experts": null, "layers": 80, "max_context": 131072, "kv_bytes_per_token_fp16": 327680, "attention_layout": "All 80 layers use full attention", "q4_k_m_weights_gib": 39.75, "q8_0_weights_gib": 69.82, "fp16_weights_gib": 131.42, "q4_k_m_8k_total_gib": 46.98, "q4_k_m_32k_total_gib": 55.23, "q4_k_m_128k_total_gib": 88.23, "q8_0_8k_total_gib": 80.05, "q8_0_32k_total_gib": 88.3, "q8_0_128k_total_gib": 121.3, "fp16_8k_total_gib": 147.81, "fp16_32k_total_gib": 156.06, "fp16_128k_total_gib": 189.06, "kv_cache_8k_gib": 2.5, "kv_cache_32k_gib": 10, "kv_cache_128k_gib": 40, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "Qwen3 8B", "hf_id": "Qwen/Qwen3-8B", "hf_url": "https://huggingface.co/Qwen/Qwen3-8B", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3-8b/", "source_checked": "2026-09-26", "total_params": 8190735360, "active_params": null, "experts": null, "layers": 36, "max_context": 40960, "kv_bytes_per_token_fp16": 147456, "attention_layout": "All 36 layers use full attention", "q4_k_m_weights_gib": 4.62, "q8_0_weights_gib": 8.1, "fp16_weights_gib": 15.26, "q4_k_m_8k_total_gib": 6.81, "q4_k_m_32k_total_gib": 10.53, "q4_k_m_128k_total_gib": null, "q8_0_8k_total_gib": 10.65, "q8_0_32k_total_gib": 14.37, "q8_0_128k_total_gib": null, "fp16_8k_total_gib": 18.52, "fp16_32k_total_gib": 22.23, "fp16_128k_total_gib": null, "kv_cache_8k_gib": 1.13, "kv_cache_32k_gib": 4.5, "kv_cache_128k_gib": null, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": true, "q8_0_32k_fits_24gib": true, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": true, "fp16_32k_fits_32gib": true, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "Qwen3 30B-A3B (MoE)", "hf_id": "Qwen/Qwen3-30B-A3B", "hf_url": "https://huggingface.co/Qwen/Qwen3-30B-A3B", "model_url": "https://modelvram.com/llm-vram-calculator/qwen3-30b-a3b/", "source_checked": "2026-09-26", "total_params": 30532122624, "active_params": 3300000000, "experts": 128, "layers": 48, "max_context": 40960, "kv_bytes_per_token_fp16": 98304, "attention_layout": "All 48 layers use full attention", "q4_k_m_weights_gib": 17.2, "q8_0_weights_gib": 30.21, "fp16_weights_gib": 56.87, "q4_k_m_8k_total_gib": 20.25, "q4_k_m_32k_total_gib": 22.72, "q4_k_m_128k_total_gib": null, "q8_0_8k_total_gib": 34.56, "q8_0_32k_total_gib": 37.03, "q8_0_128k_total_gib": null, "fp16_8k_total_gib": 63.88, "fp16_32k_total_gib": 66.36, "fp16_128k_total_gib": null, "kv_cache_8k_gib": 0.75, "kv_cache_32k_gib": 3, "kv_cache_128k_gib": null, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "gpt-oss-20b", "hf_id": "openai/gpt-oss-20b", "hf_url": "https://huggingface.co/openai/gpt-oss-20b", "model_url": "https://modelvram.com/llm-vram-calculator/gpt-oss-20b/", "source_checked": "2026-09-26", "total_params": 20914757184, "active_params": 3600000000, "experts": 32, "layers": 24, "max_context": 131072, "kv_bytes_per_token_fp16": 24576, "attention_layout": "12 of its 24 layers use full attention and 12 keep a sliding window of 128 tokens", "q4_k_m_weights_gib": 11.78, "q8_0_weights_gib": 20.7, "fp16_weights_gib": 38.96, "q4_k_m_8k_total_gib": 13.69, "q4_k_m_32k_total_gib": 14.31, "q4_k_m_128k_total_gib": 16.78, "q8_0_8k_total_gib": 23.49, "q8_0_32k_total_gib": 24.11, "q8_0_128k_total_gib": 26.58, "fp16_8k_total_gib": 43.58, "fp16_32k_total_gib": 44.2, "fp16_128k_total_gib": 46.67, "kv_cache_8k_gib": 0.21, "kv_cache_32k_gib": 0.77, "kv_cache_128k_gib": 3.02, "q4_k_m_32k_fits_16gib": true, "q4_k_m_32k_fits_24gib": true, "q4_k_m_32k_fits_32gib": true, "q4_k_m_32k_fits_48gib": true, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": true, "q8_0_32k_fits_48gib": true, "q8_0_32k_fits_80gib": true, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": true, "fp16_32k_fits_80gib": true, "note": "" }, { "model": "gpt-oss-120b", "hf_id": "openai/gpt-oss-120b", "hf_url": "https://huggingface.co/openai/gpt-oss-120b", "model_url": "https://modelvram.com/llm-vram-calculator/gpt-oss-120b/", "source_checked": "2026-09-26", "total_params": 116829156672, "active_params": 5100000000, "experts": 128, "layers": 36, "max_context": 131072, "kv_bytes_per_token_fp16": 36864, "attention_layout": "18 of its 36 layers use full attention and 18 keep a sliding window of 128 tokens", "q4_k_m_weights_gib": 65.83, "q8_0_weights_gib": 115.61, "fp16_weights_gib": 217.61, "q4_k_m_8k_total_gib": 73.25, "q4_k_m_32k_total_gib": 74.18, "q4_k_m_128k_total_gib": 77.89, "q8_0_8k_total_gib": 128, "q8_0_32k_total_gib": 128.93, "q8_0_128k_total_gib": 132.65, "fp16_8k_total_gib": 240.21, "fp16_32k_total_gib": 241.14, "fp16_128k_total_gib": 244.85, "kv_cache_8k_gib": 0.31, "kv_cache_32k_gib": 1.15, "kv_cache_128k_gib": 4.53, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": true, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" }, { "model": "DeepSeek V3 / R1 (671B)", "hf_id": "deepseek-ai/DeepSeek-V3", "hf_url": "https://huggingface.co/deepseek-ai/DeepSeek-V3", "model_url": "https://modelvram.com/llm-vram-calculator/deepseek-v3/", "source_checked": "2026-09-26", "total_params": 684531386000, "active_params": 37000000000, "experts": 256, "layers": 61, "max_context": 163840, "kv_bytes_per_token_fp16": 70272, "attention_layout": "All 61 layers use multi-head latent attention", "q4_k_m_weights_gib": 385.7, "q8_0_weights_gib": 677.36, "fp16_weights_gib": 1275.04, "q4_k_m_8k_total_gib": 425.36, "q4_k_m_32k_total_gib": 427.13, "q4_k_m_128k_total_gib": 434.21, "q8_0_8k_total_gib": 746.19, "q8_0_32k_total_gib": 747.96, "q8_0_128k_total_gib": 755.04, "fp16_8k_total_gib": 1403.63, "fp16_32k_total_gib": 1405.4, "fp16_128k_total_gib": 1412.48, "kv_cache_8k_gib": 0.54, "kv_cache_32k_gib": 2.14, "kv_cache_128k_gib": 8.58, "q4_k_m_32k_fits_16gib": false, "q4_k_m_32k_fits_24gib": false, "q4_k_m_32k_fits_32gib": false, "q4_k_m_32k_fits_48gib": false, "q4_k_m_32k_fits_80gib": false, "q8_0_32k_fits_16gib": false, "q8_0_32k_fits_24gib": false, "q8_0_32k_fits_32gib": false, "q8_0_32k_fits_48gib": false, "q8_0_32k_fits_80gib": false, "fp16_32k_fits_16gib": false, "fp16_32k_fits_24gib": false, "fp16_32k_fits_32gib": false, "fp16_32k_fits_48gib": false, "fp16_32k_fits_80gib": false, "note": "" } ] }