Mistral AI, France · Announced October 6, 2026Mistral Large 4 at a glance
- Made by
- Mistral AI, France
- Announced
- October 6, 2026
- Parameters
- 1051.6B, 49B active
- Context window
- 1M tokens
Mistral Large 4 VRAM Requirements
Mistral AI, France · Announced October 6, 2026New
Mistral Large 4 has 1051.6B parameters (49B active per token). In BF16 its weights alone take 1959 GiB; quantized to 4-bit, about 591 GiB. Each 32K-token request adds 2.14 GiB of KV cache. Cheapest way to run it today: 7 × RTX PRO 6000 Blackwell (96 GB) at $9.34 per hour.
Preview: Mistral plans to release the open weights (FP8 and FP4) by the end of October 2026. Until then the layer layout and the FP8 size are estimates based on Mistral Large 3; we update them when the official config ships.
Size your own setup in the calculator →VRAM by precision and context length
Weights plus the KV cache of one request (BF16 cache), in GiB. Add 1 to 3 GiB for the inference engine itself.
| Weights format | Weights only | + 8K context | + 32K context | + 128K context | + 1M context |
|---|---|---|---|---|---|
| FP8 (native) | 989 | 989 | 991 | 997 | 1057 |
| BF16 / FP16 | 1959 | 1959 | 1961 | 1967 | 2027 |
| FP8 / INT8 | 981 | 982 | 983 | 990 | 1050 |
| 4-bit (GGUF Q4_K_M) | 591 | 592 | 594 | 600 | 660 |
How many GPUs to run Mistral Large 4
Fewest GPUs that fit the weights plus one 8K-token request. Data-center GPUs use vLLM (90% of memory usable), the others llama.cpp. Several GPUs split the model with tensor or pipeline parallelism.
| GPU | FP8 (native) | BF16 / FP16 | FP8 / INT8 | 4-bit (GGUF Q4_K_M) |
|---|---|---|---|---|
| GeForce RTX 4090 24 GB | 64 (several servers) | 128 (several servers) | 64 (several servers) | 32 (several servers) |
| GeForce RTX 5090 32 GB | 48 (several servers) | 96 (several servers) | 48 (several servers) | 24 (several servers) |
| RTX PRO 6000 Blackwell 96 GB | 16 (several servers) | 24 (several servers) | 16 (several servers) | 7 |
| Mac M5 Max 128 GB 96 GB usable | 16 (several servers) | 24 (several servers) | 16 (several servers) | 7 |
| NVIDIA H100 80 GB | 16 (several servers) | 32 (several servers) | 16 (several servers) | 16 (several servers) |
| NVIDIA H200 141 GB | 16 (several servers) | 16 (several servers) | 8 | 5 |
| NVIDIA B200 (HGX) 180 GB | 7 | 16 (several servers) | 7 | 4 |
Cheapest way to run Mistral Large 4 today
For each GPU: the fewest cards that fit Mistral Large 4 with one 8K-token request, the most faithful weight format at that count, and the cheapest on-demand price (2026-10-10).
| Setup | Per hour | Weights | Provider | Per month | Rent |
|---|---|---|---|---|---|
| 7 × RTX PRO 6000 Blackwell 96 GB | $9.34 | 4-bit (GGUF Q4_K_M) | Vast.ai | $6,816.74 | Rent 7 × RTX PRO 6000 Blackwell on Vast.ai (opens in a new tab) → |
| 4 × AMD MI300X 192 GB | $9.56 | 4-bit (GGUF Q4_K_M) | RunPod | $6,978.80 | Rent 4 × AMD MI300X on RunPod (opens in a new tab) → |
| 3 × AMD MI355X / MI350X 288 GB | $16.47 | 4-bit (GGUF Q4_K_M) | RunPod | $12,023.10 | Rent 3 × AMD MI355X / MI350X on RunPod (opens in a new tab) → |
| 8 × NVIDIA H100 NVL 94 GB | $19.74 | 4-bit (GGUF Q4_K_M) | Vast.ai | $14,413.12 | Rent 8 × NVIDIA H100 NVL on Vast.ai (opens in a new tab) → |
| 5 × NVIDIA H200 141 GB | $23.04 | 4-bit (GGUF Q4_K_M) | Vast.ai | $16,815.55 | Rent 5 × NVIDIA H200 on Vast.ai (opens in a new tab) → |
| 4 × NVIDIA B200 (HGX) 180 GB | $25.00 | 4-bit (GGUF Q4_K_M) | Vast.ai | $18,252.92 | Rent 4 × NVIDIA B200 (HGX) on Vast.ai (opens in a new tab) → |
| 3 × NVIDIA B300 / GB300 288 GB | $26.97 | 4-bit (GGUF Q4_K_M) | RunPod | $19,688.10 | Rent 3 × NVIDIA B300 / GB300 on RunPod (opens in a new tab) → |
On-demand prices from RunPod, Vast.ai, Verda and Azure, checked every hour. Serving many users needs more KV cache, so more memory: size it in the calculator. Some links are affiliate links: StudioTV may earn a commission, at no extra cost to you.
KV cache: how memory grows with context
Architecture: Granular MoE · 1.6B vision encoder · MLA (512 + 64 latent) on 61 layers, assumed from Mistral Large 3. Its cache stays compact because MLA stores one compressed latent per token and layer instead of full keys and values. Only 49B parameters are active per token, but all 1052B must sit in memory: VRAM depends on the total, speed on the active part.
| Context per request | BF16 cache | FP8 cache |
|---|---|---|
| 8K tokens | 0.54 GiB | 0.27 GiB |
| 32K tokens | 2.14 GiB | 1.07 GiB |
| 128K tokens | 8.58 GiB | 4.29 GiB |
| 1M tokens | 68.6 GiB | 34.3 GiB |
Mistral Large 4 VRAM FAQ
How much VRAM does Mistral Large 4 need?
In BF16 the weights alone take 1959 GiB (2103 GB). In FP8 that is 981 GiB, and about 591 GiB with 4-bit quantization (Q4_K_M); the official FP8 checkpoint should take about 989 GiB. Each request then adds KV cache: 0.54 GiB at 8K tokens and 2.14 GiB at 32K (BF16 cache).
Can Mistral Large 4 run on a single RTX 4090 (24 GB)?
No. Even in 4-bit it needs 32 (several servers) RTX 4090s. In 4-bit it needs 16 (several servers) H100 GPUs.
What is the cheapest way to run Mistral Large 4?
On 2026-10-10, the cheapest on-demand setup is 7 × RTX PRO 6000 Blackwell (96 GB) with 4-bit (GGUF Q4_K_M) weights on Vast.ai, at $9.34 per hour (about $6,816.74 per month). Next: 4 × AMD MI300X (192 GB) with 4-bit (GGUF Q4_K_M) weights on RunPod, at $9.56 per hour. Sized for one 8K-token request; prices are checked every hour.
How many H100 GPUs does Mistral Large 4 need?
With one request and an 8K-token context: 32 (several servers) in BF16, 16 (several servers) in FP8 and 16 (several servers) in 4-bit (vLLM, 90% of memory usable). Serving many users at once needs more memory for their KV caches: the calculator sizes that for you.
Other models
Same family (large moe): MiniMax M2.7 · DeepSeek R1 · Kimi K2.6 · GLM-5.3 · GLM-5.3-Flash · DeepSeek V4 Flash · DeepSeek V4 Pro · DeepSeek V4.1 Flash · MiMo-V2.6 Pro · MiMo-V2.6 Flash · Kimi K3 · DeepSeek V4 Flash Vision Exp · Qwen3.8 2.4T-A95B · Ornith 1.5 397B · Atria Dawn Preview · Intern S2 397B
All models: Llama 3.1 8B · Mistral Small 3.2 24B · Qwen3 32B · Llama 3.3 70B · Llama 3.1 405B · Qwen3.5 4B · Qwen3.5 9B · Qwen3.8 27B · Qwen3.6 35B-A3B · Qwen3.5 122B-A10B · Qwen3.8-Flash-Next · LFM2.5 8B-A1B · LFM2 24B-A2B · Gemma 4 12B · Gemma 4 31B · Gemma 4 26B-A4B · gpt-oss 20B · gpt-oss 120B · Llama 4 Scout 17B-16E · Llama 4 Maverick 17B-128E · Kolibri-1 · Nemotron 3 Nano 30B-A3B · Nemotron 3 Super 120B-A12B · MiniMax M2.7 · DeepSeek R1 · Kimi K2.6 · GLM-5.3 · GLM-5.3-Flash · DeepSeek V4 Flash · DeepSeek V4 Pro · DeepSeek V4.1 Flash · MiMo-V2.6 Pro · MiMo-V2.6 Flash · Kimi K3 · Qwen3-Coder-Next 80B-A3B · Qwen3-Coder 30B-A3B · Qwen3-Coder 480B-A35B · DeepSeek V4 Flash Vision Exp · Qwen3.8 2.4T-A95B · Ornith 1.5 397B · Ornith 1.5 35B-A3B · Ornith 1.5 9B · Atria Dawn Preview · Intern S2 397B · Clef · Clef Flash · D1 3B · Humanizer · JEV 27B VL · Mellum2.1 12B-A2.5B · LightOnOCR-3 4B · Spark-X2.5 4B · Agnes-3.0-Qwen
Estimates, not guarantees: computed from the official config.json with the same engine as the LLM VRAM Calculator. Real usage depends on your engine version and settings.