HolyGrAIL

All tasks / Models

Devstral 2 (2512)

Mistral AI's open-weights coding-agent model line; Mistral says Medium 3.5 now supersedes Devstral on its benchmarks.

  • Mistral AI
  • Open weights
  • Released 9 Dec 2025

Official page

Where it ranks

  • C
    Coding
    Models · Local · 1,197 on LMArena WebDev / Code Arena (overall)

Facts

Vendor
Mistral AI
Family
Devstral
Released
9 Dec 2025
Weights
Open: download and run it yourself
License
Modified MIT (per LMArena; HF lists 'other')
Parameters
123B
GPU memory to run locally
~72 GB (4-bit ≈ 123 B × 0.57 GB + 1.5 GB ≈ 72 GB, estimate.)
API price
$0.4 in / $2 out per 1M tokens
Access
Free (open weights)
Last verified
18 Oct 2026

Price from openrouter.ai, 18 Oct 2026.

Benchmark results

BenchmarkResultEntry on the leaderboardAs ofSource
LMArena WebDev / Code Arena (overall)1,197Devstral 2
From LMArena's official leaderboard-dataset (webdev = Code Arena, split latest, category overall, published 2026-10-18), which mirrors https://arena.ai/leaderboard/webdev. Rank 140; 1182 votes; 95% CI 1175.8-1217.3.
18 Oct 2026huggingface.co
SWE-bench Verified (vendor-reported)72.2%Devstral 2
vendor-reported in model card, harness: not stated
18 Oct 2026huggingface.co

Where you can use it

Earlier versions

  • Devstral Medium (2507) · —

Sources