Mistral Large 4: Medium 3.5 comparison no longer gives an unsourced RAM figure; it points to that checker instead

Mistral AI released Mistral Large 4, a 1.05-trillion-parameter mixture-of-experts model with 52B parameters active per token, as a public preview on October 6, 2026, with open weights announced for the end of October and October 31 as the current ETA on Mistral's Hugging Face page. According to memory estimates derived from the published parameter count, the weights alone require about 2.1 TB at BF16, 1.05 TB at FP8, roughly 0.56–0.63 TB at 4-bit and about 0.28–0.51 TB at 2–3-bit GGUF quants, meaning an 8-GPU server at FP8 or a 512 GB Mac Studio for the smallest quants, and it will not fit an RTX 4090/5090. The figures remain estimates until an official model card lists file sizes, and runtime support is unknown until the weights ship.

Mistral Large 4 hardware requirements: VRAM, GPUs & quantization - Page updated - Sources checked - Specs from Mistral's docs and announcement - Estimates labeled until the model card ships - Jump to the checker ↓ calculator Mistral Large 4 "le Chonk", Mistral's own nickname for it is a 1.05-trillion-parameter mixture-of-experts model with 52B parameters active per token 49B not counting embeddings and output layers . Mistral released it as a public preview on October 6, 2026; the open weights are announced for the end of October, with October 31 as the current ETA on Mistral's Hugging Face page. Here is how much memory it takes to run it yourself. Short answer: the weights alone take about 2.1 TB at BF16 , 1.05 TB at FP8 , ~0.56–0.63 TB at 4-bit and ~0.28–0.51 TB at 2–3-bit GGUF quants . Only 52B parameters are active per token 49B not counting embeddings and output layers , but all 1.05T must be held in memory. By capacity alone that is an 8-GPU server at FP8, or a 512 GB Mac Studio for the smallest quants; it will not fit an RTX 4090/5090. These are memory estimates; runtime support is unknown until the weights ship. Weights: announced for the end of October 2026. Every figure below is an estimate from the published parameter count until an official model card lists file sizes. How much VRAM does Mistral Large 4 need? Weights = parameters × bits per weight ÷ 8. "Block" is the storage rate of the format itself; GGUF " K"/" M" files keep some tensors at higher precision, so we assume a slightly higher whole-model average the number used for the size . Device counts add 10% for KV cache and buffers and assume a usable budget per device H200 131 GB, B200 168 GB, H100 74 GB, 512 GB Mac 460 GB . They are theoretical capacity minimums, not validated deployments. | Format | Block bits | Assumed avg | Weights | H200 | B200 | H100 | Mac 512 GB | |---|---|---|---|---|---|---|---| | BF16 | 16 | 16 | ~2.1 TB | 18 | 14 | 32 | 6 | | FP8 | 8 | 8 | ~1.05 TB | 9 | 7 | 16 | 3 | | Q8 0 | 8.5 | 8.5 | ~1.12 TB | 10 | 8 | 17 | 3 | | Q6 K | 6.56 | ~6.6 | ~866 GB | 8 | 6 | 13 | 3 | | Q4 K M | 4.5 | ~4.8 | ~630 GB | 6 | 5 | 10 | 2 | | NVFP4 | 4.5 | 4.5 | ~591 GB | 5 | 4 | 9 | 2 | | MXFP4 | 4.25 | 4.25 | ~558 GB | 5 | 4 | 9 | 2 | | Q3 K M | 3.44 | ~3.9 | ~512 GB | 5 | 4 | 8 | 2 | | Q2 K | 2.63 | ~3.0 | ~394 GB | 4 | 3 | 6 | 1 | | IQ2 XXS | 2.06 | ~2.1 | ~276 GB | 3 | 2 | 5 | 1 | | IQ1 S | 1.56 | ~1.6 | ~210 GB | 2 | 2 | 4 | 1 | Block bits: llama.cpp ggml-common.h https://github.com/ggml-org/llama.cpp/blob/master/ggml/src/ggml-common.h ; NVFP4 stores an 8-bit scale per 16 values, MXFP4 per 32. Whole-model averages for this architecture are unknown until real files exist; set your own in the calculator. Below ~3 bits, quality on a model this size is untested. Mistral Large 4 VRAM & memory calculator Open in a full page ↗ https://tools.localrigcheck.com/mistral-large-4/ Loading private checker… - Developer - Mistral AI Official src-2 - Total parameters - 1.05 trillion Mistral docs Official src-1 - Active parameters - 52B per token Official src-1 - Context - 1M tokens Mistral docs, API model page Official src-1 - Weights - open weights announced for the end of October 2026 Official src-2 - API - mistral-large-4-0 Official src-1 - Ollama - only the cloud tag mistral-large-4:cloud Official src-4 Can you run Mistral Large 4 on a consumer GPU? Not on GPUs alone: even IQ1 S ~210 GB is more than seven RTX 5090s. If runtimes add support once the weights ship, the home-scale options by memory capacity are: - One 512 GB Mac Studio M3 Ultra : room for Q2 K, IQ2 XXS or IQ1 S under our assumptions. Q3 K M ~512 GB + overhead already needs two machines. Requires MLX or llama.cpp support for the model. - Two clustered 512 GB Mac Studios : room for 4-bit ~560–630 GB , via clustering software such as exo or llama.cpp RPC. - A server with 512 GB–1 TB of RAM and one large GPU : llama.cpp-style MoE offload keeps shared layers on the GPU and experts in RAM. The idealized batch-one estimate on 8-channel DDR5 ~300 GB/s is ~10 tokens/s at Q4 K M and ~15 at Q2 K under those assumptions; GPU-resident layers, batching or speculative decoding change it. Mixture-of-experts helps speed, not memory: only 52B parameters work per token, but the router can pick any expert, so all of them must be loaded. Self-host vs. the Mistral API The comparison is inside the private calculator. Open the cost calculator → https://tools.localrigcheck.com/mistral-large-4/ cost . Mistral Large 4 specs | Total parameters | 1.05 trillion Mistral docs . The docs also list a 1.6B vision encoder without saying whether the 1.05T includes it; our estimates use 1.05T. If it comes on top, it adds about 3.2 GB at BF16, which changes none of the device counts in the VRAM table above 10% overhead . | |---|---| | Active parameters | 52B per token mixture of experts; Mistral docs and announcement . Mistral's Hugging Face page splits it: 49B per token, 52B including embeddings and output layers. Our speed estimate uses 52B. | | Type | Hybrid instruct + reasoning, natively multimodal text and images | | Languages | 160+ | | Nicknames | ML4, le Chonk both used in Mistral's announcement | | Release | Public preview October 6, 2026; open weights announced for the end of October 2026 Mistral's Hugging Face page: October 31, 2026, "Current ETA" | | API | mistral-large-4-0 , $1.36 / $4.18 per million input / output tokens list price; Mistral's docs showed a launch sale at half that on October 8, 2026 | | Context length | 1M tokens Mistral docs, API model page | | License, file formats | Not stated in the announcement or docs checked October 8, 2026 | FAQ How much VRAM does Mistral Large 4 need? By our estimate about 2.1 TB at BF16, 1.05 TB at FP8 and roughly 560–630 GB at 4-bit for the weights alone, plus memory for the KV cache. At FP8 that is the capacity of an 8-GPU server such as 8× B200 or 8× MI300X. Official file sizes are not published yet. Can Mistral Large 4 run on an RTX 4090 or RTX 5090? No. Even a ~1.6-bit quant is about 210 GB, far beyond one 24–32 GB card. A server with enough system RAM could offload the experts to the CPU if a runtime supports the model, with speed limited by RAM bandwidth. Can I run Mistral Large 4 with a 24 GB GPU and 128 GB of RAM? Not by our capacity estimate. 24 GB of VRAM plus 128 GB of RAM is 152 GB, and even the smallest quant in the table above, IQ1 S, is about 210 GB before the KV cache. Offloading the experts to system RAM does not shrink them: the router can pick any expert, so they all have to fit somewhere. The weights are not downloadable yet, so local runtime support and setups that stream experts from an SSD are untested; we know of no measurement. On such a PC, Mistral's API or Ollama's cloud model is the way to use it. A dense model that does fit there is Mistral Medium 3.5 128B : by our estimate its 4-bit GGUF runs on a 24 GB GPU with most of its layers in system RAM, slowly the Medium 3.5 checker shows the split for your RAM . See the Mistral Medium 3.5 checker https://localrigcheck.com/mistral-medium-3-5/ . Can I run Mistral Large 4 on a Mac Studio? By memory capacity, a 512 GB M3 Ultra can hold Q2 K, IQ2 XXS or IQ1 S quants; Q3 K M and 4-bit need two clustered machines. That also requires MLX or llama.cpp support for the model, which is not confirmed yet. Why does a 52B-active model need memory for 1.05T parameters? In a mixture-of-experts model a router picks different experts for every token, so all experts must be loaded. Active parameters decide speed bytes read per token ; total parameters decide memory. What is Le Chonk Mistral Chonk ? Le Chonk is Mistral AI's own nickname for Mistral Large 4: the announcement says "Unofficially ML4, very officially: le Chonk." It is the same 1.05T-parameter model with 52B active parameters, so every memory figure on this page applies to it. When will Mistral Large 4 weights be released? Mistral's announcement says it will release the weights "by the end of the month" October 2026 , and its Hugging Face page for the model, mistralai/Mistral-Large-4.0-1T05-A52B , lists October 31, 2026 as the current ETA checked October 8, 2026 . Community GGUF, MLX or local Ollama builds need both the weights and runtime support; no date for those is confirmed. Is Mistral Large 4 on Ollama? Only as a cloud model so far. Ollama's library lists mistral-large-4:cloud checked October 8, 2026 , which runs on Ollama's servers, not on your hardware, so none of the memory figures on this page apply to it. Ollama lists no local tags, and no downloadable weights are out yet both checked October 8, 2026 ; whether Ollama or llama.cpp will support the architecture locally is not known until they ship. Mistral's own API is the other way to use it today. Sources - Mistral docs: Mistral Large 4 model page https://docs.mistral.ai/models/mistral-large-4 1.05T total and 52B active parameters, vision encoder, context, API price - Mistral AI: Mistral Large 4 announcement https://mistral.ai/news/mistral-large-4 parameters, pricing, release plan, the "le Chonk" nickname - Hugging Face: mistralai/Mistral-Large-4.0-1T05-A52B https://huggingface.co/mistralai/Mistral-Large-4.0-1T05-A52B placeholder without files; 49B per token, 52B including embeddings and output layers; October 31 ETA; checked October 8, 2026 - Ollama library: mistral-large-4 tags https://ollama.com/library/mistral-large-4/tags only the cloud tag mistral-large-4:cloud ; checked October 8, 2026 - llama.cpp ggml-common.h https://github.com/ggml-org/llama.cpp/blob/master/ggml/src/ggml-common.h GGUF block formats - MLX LM README https://github.com/ml-explore/mlx-lm/blob/main/README.md raising the macOS GPU memory limit - Apple: Mac Studio 2025 tech specs https://support.apple.com/en-au/122211 819 GB/s, 512 GB - Device budgets, overhead and whole-model averages are RigCheck assumptions, stated above.