Skip to content
English

Switch theme

// Mistral / Magistral / Devstral · Text LLM

Ministral-3-8B-Reasoning-2512: VRAM & local deployment

8.92B parameters. About 7.6 GB at 4-bit and 21.3 GB at BF16 with one 8K-token request.

released 2025-10-31licence: Commercial OKmemory figures are estimates

01Specs

PublisherMistral AI · mistralai
Family / seriesMistral / Magistral / Devstral · Ministral-3-Reasoning
TypeText LLM
Parameters8.92B Source: counted from official safetensors weights
Context length256K tokens (262,144)max_position_embeddings in config.json
Architecture34 layers · GQA
Released weightsBF16 · 35.7 GBdownload
LicenceApache-2.0 · Commercial OKPermissive licence (Apache-2.0 / MIT / CC-BY / OpenMDW): commercial use allowed.
Released2025-10-31 (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.

PrecisionWeightsTotal @ 8KTotal @ 32KTotal @ 128KSmallest GPUMac
BF16 / FP16 (original)17.8 GB 21.3 GB24.7 GB38.4 GB RTX 4090 24GBMac 36GB
FP8 / INT8 / Q8_09.5 GB 12.1 GB15.5 GB29.2 GB 16GB GPUMac 24GB
4-bit (Q4_K_M / AWQ / GPTQ-Int4)5.4 GB 7.6 GB11.0 GB24.7 GB 12GB GPUMac 16GB

KV cache per request: 8K 1.1 GB · 32K 4.6 GB · 128K 18.3 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.

Which inference engine to choose →

04Official repositories & quantised builds

2 official repo(s) from mistralai.

05Other Mistral / Magistral / Devstral models

All 35 →
ModelParams4-bit totalReleased
Ministral-3-3B-Reasoning-2512 · same series4.25B4.2 GB2025-10-31
Ministral-3-14B-Reasoning-2512 · same series13.9B11.1 GB2025-10-31
LIDstral-Arabic——2026-10-08
Voxtral-Mini-4B-Realtime-Arabic4.43B3.5 GB2026-10-08
Shieldstral-1.0-3B3.85B3.9 GB2026-07-16
Mistral-Medium-3.5-128B127.7B88.6 GB2026-03-31
Mistral-Small-4-119B-2603119B / 6.5B80.0 GB2026-01-23
Voxtral-Mini-4B-Realtime-26024.43B3.5 GB2026-01-21
Devstral-2-123B-Instruct-2512125B86.8 GB2025-11-28
Mistral-Large-3-675B-Instruct-2512675B / 41B451 GB2025-11-28

06Sources

Data checked 2026-07-15. Parameters, context and licence are copied from the official repository; memory figures are our estimates.

Data updated: Self-host open models(UTC+8)