// Mistral / Magistral / Devstral · Text LLM
Mistral-Medium-3.5-128B: VRAM & local deployment
127.7B parameters. About 88.6 GB at 4-bit and 284 GB at BF16 with one 8K-token request.
released 2026-03-31licence: Custom licence – read itmemory figures are estimates
01Specs
| Publisher | Mistral AI · mistralai |
|---|---|
| Family / series | Mistral / Magistral / Devstral · Mistral-Medium-3.5 |
| Type | Text LLM |
| Parameters | 127.7B Source: counted from official safetensors weights |
| Context length | 256K tokens (262,144)max_position_embeddings in config.json |
| Architecture | 88 layers · GQA |
| Released weights | FP8 (E4M3) · 267 GBdownload |
| Licence | other · Custom licence – read itThe publisher uses its own licence; we have not classified it. Read the licence text. |
| Released | 2026-03-31 (Hugging Face repo created) |
02How much VRAM it needs (estimate)
Total = weights + 10% + 0.5GB overhead + KV cache for one request (FP16 KV). More concurrent users multiply the KV column.
| Precision | Weights | Total @ 8K | Total @ 32K | Total @ 128K | Smallest GPU | Mac |
|---|---|---|---|---|---|---|
| BF16 / FP16 (original) | 255 GB | 284 GB | 293 GB | 329 GB | 4 × RTX PRO 6000 (384GB) | Mac 512GB |
| FP8 / INT8 / Q8_0 | 136 GB | 153 GB | 162 GB | 197 GB | 4 × RTX PRO 6000 (384GB) | Mac 256GB |
| 4-bit (Q4_K_M / AWQ / GPTQ-Int4) | 77.4 GB | 88.6 GB | 97.5 GB | 133 GB | H200 141GB | Mac 256GB |
KV cache per request: 8K 3.0 GB · 32K 11.8 GB · 128K 47.2 GB
Bits per weight: BF16 16, FP8 8.5, 4-bit 4.85. “Fits” = total ≤ 90% of memory; Mac GPU can use ~2/3 of unified memory (≤36GB) or ~3/4 (larger). For MoE, all experts must be in memory, so size by total parameters. Method →
03How to run it
Only runtimes the publisher or the runtime’s official library names for this model.
- vLLM — mentioned in the official model card
- SGLang — mentioned in the official model card
- llama.cpp — mentioned in the official model card
- LM Studio — mentioned in the official model card
- Transformers — mentioned in the official model card
- Ollama — in the official Ollama library · mistral-medium-3.5:128b
ollama run mistral-medium-3.5:128b04Official repositories & quantised builds
1 official repo(s) from mistralai.
05Other Mistral / Magistral / Devstral models
All 35 →| Model | Params | 4-bit total | Released |
|---|---|---|---|
| LIDstral-Arabic | — | — | 2026-10-08 |
| Voxtral-Mini-4B-Realtime-Arabic | 4.43B | 3.5 GB | 2026-10-08 |
| Shieldstral-1.0-3B | 3.85B | 3.9 GB | 2026-07-16 |
| Mistral-Small-4-119B-2603 | 119B / 6.5B | 80.0 GB | 2026-01-23 |
| Voxtral-Mini-4B-Realtime-2602 | 4.43B | 3.5 GB | 2026-01-21 |
| Devstral-2-123B-Instruct-2512 | 125B | 86.8 GB | 2025-11-28 |
| Mistral-Large-3-675B-Instruct-2512 | 675B / 41B | 451 GB | 2025-11-28 |
| Voxtral-4B-TTS-2603 | 4B | 3.2 GB | 2025-11-17 |
06Sources
- Hugging Face · mistralai/Mistral-Medium-3.5-128B
- config.json
- License
- Ollama library · mistral-medium-3.5
Data checked 2026-07-15. Parameters, context and licence are copied from the official repository; memory figures are our estimates.