https://modelvram.com/changelog/ New calculators, model pages, data and fixes on ModelVRAM. en Wed, 30 Sep 2026 12:00:00 GMT https://modelvram.com/accuracy/#prompt-speed https://modelvram.com/changelog/#accuracy-prompt-speed Wed, 30 Sep 2026 12:00:00 GMT The accuracy page now also lists all 32 llama-bench runs behind the time-to-first-token estimate, with the error per hardware family, and the ComfyUI load logs behind the image and video file sizes. https://modelvram.com/best-llm-for-vram/ https://modelvram.com/changelog/#vendor-reported-benchmarks Wed, 30 Sep 2026 12:00:00 GMT GPQA, MMLU-Pro and LiveCodeBench scores that makers published in their own model cards, linked to the source: on model pages, as a GPQA column in the per-size lists, and as a y-axis option on the chart. https://modelvram.com/llm-vram-calculator/bespoke-nimble-9b/ https://modelvram.com/changelog/#bespoke-nimble-9b Wed, 30 Sep 2026 12:00:00 GMT VRAM per quant and context for Bespoke Nimble 9B, with its GGUF files and Ollama tags. https://modelvram.com/llm-vram-calculator/fit-context-regression/ https://modelvram.com/changelog/#fit-context-regression Wed, 30 Sep 2026 12:00:00 GMT Why llama.cpp builds b10999–b11200 picked contexts such as 828,160 tokens, how --fit sizes context, slots and layers now, and the flags to set. https://modelvram.com/llm-vram-calculator/gpu/ https://modelvram.com/changelog/#gpu-index Wed, 30 Sep 2026 12:00:00 GMT Every GPU, Mac, CPU-only PC and multi-GPU setup on one page, with how many models each runs. https://modelvram.com/llm-vram-calculator/ https://modelvram.com/changelog/#cloud-gpu-card Wed, 30 Sep 2026 12:00:00 GMT When a model does not fit a local card, the calculators name the memory it needs and the cloud GPU class that holds it, with plain, unsponsored links. https://modelvram.com/llm-vram-calculator/ https://modelvram.com/changelog/#ollama-names Wed, 30 Sep 2026 12:00:00 GMT Type an Ollama library name such as qwen3:30b-a3b into the LLM VRAM calculator to load that model. https://modelvram.com/llm-vram-calculator/lensvlm-9b/ https://modelvram.com/changelog/#lensvlm-9b Wed, 30 Sep 2026 12:00:00 GMT VRAM for Apple’s LensVLM-9B per quant and context, its GGUF files and mmproj, and what its visual text compression saves in KV cache. https://modelvram.com/laya-vram-requirements/ https://modelvram.com/changelog/#decision-latency Wed, 30 Sep 2026 12:00:00 GMT How long one decision takes and how many per second a GPU or CPU handles, on the Laya and Jev alternatives pages. https://modelvram.com/ternary-bonsai-2-27b-vram/ https://modelvram.com/changelog/#bonsai-2-27b Wed, 30 Sep 2026 12:00:00 GMT Real file sizes of the PTQ1_0, PQ2_0 and MLX files and the longest context on 8, 12 and 16 GB GPUs. https://modelvram.com/llm-vram-calculator/#fit-preview https://modelvram.com/changelog/#fit-preview Tue, 29 Sep 2026 12:00:00 GMT The LLM VRAM calculator shows what llama-server’s automatic --fit would pick for context, slots and GPU layers, and the explicit flags instead. https://modelvram.com/best-llm-for-vram/#chart https://modelvram.com/changelog/#vram-scatter Tue, 29 Sep 2026 12:00:00 GMT A scatter plot of all tracked models by VRAM at Q4_K_M against parameters, with 8–80 GB card lines. https://modelvram.com/reports/local-llm-vram-2026/ https://modelvram.com/changelog/#report-2026 Tue, 29 Sep 2026 12:00:00 GMT The largest open model per memory tier and the 32K KV cache ranking, with CSV downloads under CC BY 4.0. https://modelvram.com/image-video-vram-calculator/ https://modelvram.com/changelog/#image-video Tue, 29 Sep 2026 12:00:00 GMT Peak VRAM and system RAM for FLUX.2, Wan 2.2, LTX-2 and Qwen-Image 2.1 in ComfyUI, checked against public ComfyUI runs. https://modelvram.com/jev-local-alternatives/ https://modelvram.com/changelog/#jev-alternatives Tue, 29 Sep 2026 12:00:00 GMT Open models that do the API-only Jev’s job on your own GPU or CPU, with file sizes and VRAM. https://modelvram.com/laya-vram-requirements/ https://modelvram.com/changelog/#laya Tue, 29 Sep 2026 12:00:00 GMT Real file sizes of Laya’s checkpoints in PyTorch, ggmlc and GGUF, and whether a 16 GB or 24 GB GPU or a CPU runs it. https://modelvram.com/accuracy/ https://modelvram.com/changelog/#accuracy Tue, 29 Sep 2026 12:00:00 GMT The estimates next to 20 public llama.cpp and vLLM measurements, with the error of each. https://modelvram.com/llm-vram-calculator/kv-cache/ https://modelvram.com/changelog/#kv-cache-llama-cpp Tue, 29 Sep 2026 12:00:00 GMT Gemma 4’s global layers and sliding-window caches now match llama.cpp’s logs to the byte. https://modelvram.com/data/ https://modelvram.com/changelog/#open-data Tue, 29 Sep 2026 12:00:00 GMT Every model’s VRAM estimates as CSV and JSON under CC BY 4.0, and the models that fit 8 to 96 GB. https://modelvram.com/what-can-i-run/ https://modelvram.com/changelog/#what-can-i-run Tue, 29 Sep 2026 12:00:00 GMT Detects your GPU in the browser and lists the local LLMs it runs, with the best quantization and speed. https://modelvram.com/vllm-memory-calculator/ https://modelvram.com/changelog/#vllm Tue, 29 Sep 2026 12:00:00 GMT The KV cache pool vLLM allocates, in tokens, and how many requests fit at once, checked against startup logs. https://modelvram.com/fine-tuning-vram-calculator/ https://modelvram.com/changelog/#fine-tuning Tue, 29 Sep 2026 12:00:00 GMT GPU memory for full fine-tuning, LoRA and QLoRA in Transformers or Unsloth, checked against 15 published runs. https://modelvram.com/moe-offload-calculator/ https://modelvram.com/changelog/#moe-offload Mon, 28 Sep 2026 12:00:00 GMT The smallest llama.cpp --n-cpu-moe that fits your GPU, from real GGUF tensor sizes. https://modelvram.com/qwen-image-2-1-vram-calculator/ https://modelvram.com/changelog/#qwen-image Mon, 28 Sep 2026 12:00:00 GMT Peak VRAM for every DiT, text encoder and VAE combination of Qwen-Image 2.1, with real measurements. https://modelvram.com/llm-vram-calculator/ https://modelvram.com/changelog/#launch Sat, 26 Sep 2026 12:00:00 GMT The first two tools: how much GPU memory a model needs from its Hugging Face files, and how fast it writes on your hardware.