Autor: Jakub Rusinowski · Ostatnia aktualizacja: 5 kwietnia 2026
Model library → Devstral → Devstral-2 123B
The flagship coding agent, and dense — Mistral deliberately did not build this one sparse, so all 123B stream per token and throughput scales accordingly. Scores 71.6% on SWE-bench Verified, the best open-source coding-agent result at release. Requires about 75 GB at Q4_K_M, so a dual-GPU workstation. Apache 2.0.
Devstral-2 123B needs about 75 GB of VRAM at Q4_K_M — quantized weights plus framework overhead, before any KV cache. On Apple Silicon that figure comes out of unified memory.
| Parameters | 123 Billion (dense) |
| Context window | 262,144 |
| Architecture | Dense Transformer |
| Provider | Mistral AI |
| Licence | Apache 2.0 |
| Specified at | Q4_K_M |
| System RAM | 128 GB |
| Record updated | 2026-04-05 |
Apache-2.0 — commercial use permitted. Commercial use permitted. No usage restrictions beyond attribution.
Modelled on a reference NVIDIA RTX 4090 (24 GB). Weights plus framework overhead only — this model publishes no architecture we can read, so no KV cache is included. A real session needs more; the figure is a floor, not a target. Speed figures are ESTIMATES from the memory-bandwidth roofline described on the methodology page, not benchmarks we ran — rows marked measured come from published or reader-submitted runs. VRAM here includes the KV cache, so it reads higher than the headline figure above, which does not.
| Quant | Bits/weight | Weights | VRAM needed | Est. speed | Fit on 24 GB |
|---|---|---|---|---|---|
| Q2_K | 2.63 | 40.4 GB | 41.2 GB | ~3 tok/s (est.) | Offloads to system RAM (slow) |
| Q3_K_M | 3.41 | 52.4 GB | 53.2 GB | ~2 tok/s (est.) | Offloads to system RAM (slow) |
| Q4_K_M | 4.83 | 74.3 GB | 75.1 GB | — | Won't fit |
| Q5_K_M | 5.67 | 87.2 GB | 88 GB | — | Won't fit |
| Q6_K | 6.56 | 100.9 GB | 101.7 GB | — | Won't fit |
| Q8_0 | 8.50 | 130.7 GB | 131.5 GB | — | Won't fit |
| F16 | 16.00 | 246 GB | 246.8 GB | — | Won't fit |
Want to set your own context length and KV-cache quantization? Use the interactive VRAM calculator.
or compare on Vast.ai from $0.77/hr (typical low · varies)
As an Amazon Associate we earn from qualifying purchases. Cloud GPU links are referral links — we may earn a commission at no extra cost to you.
The cheapest catalogued GPU that runs Devstral-2 123B is the AMD Ryzen AI Max+ 395 (96 GB).
Install Ollama, then run:
ollama run devstral:123bWeights on Hugging Face: mistralai/Devstral-2-123B-Instruct-2512.
Quality scores as published by the model's authors or an independent evaluator — not throughput, and not measured by us.
| Benchmark | Score | Provenance |
|---|---|---|
| SWE-bench Verified | 71.6 / 100 % | reported |
| HumanEval | 91.2 / 100 % | reported |
| LiveCodeBench v6 | 67.8 / 100 % | reported |
Best for: software engineering, agentic coding, repo level, test driven dev.
← All Devstral models | VRAM calculator | Check your own hardware