AI Potluck
Model components / Fine-tuned / chat models

Devstral

Mistral AI

Mistral's dedicated open-weight agentic coding model family, purpose-built for software engineering tasks including SWE-bench. The Small-2 (24B) instruct variant delivers competitive coding performance in a deployable size with 321K downloads. The larger Devstral-2-123B pushes frontier-class performance under a research license.

Devstral Small 1.1 (mistralai/Devstral-Small-2507), 24B dense agentic coding model, 128k context, built with All Hands AI; updated from Devstral Small 1.0 (2505). Agentic/coding post-trained SKU. Verified live on HF June 2026.

Openness

3 high confidence
3.0
weights
open(Apache-2.0, on HF)
data
closed
code
partial(inference/serving
license
Apache-2.0(OSI permissive, no use restrictions)

Apache-2.0 open weights, but training/post-training data and pipeline undisclosed -> open_weights, not open_source.

Adoption

2 medium confidence
2.0

~37k downloads last month for the Small-2507 SKU on HF; specialist coding-agent model with moderate uptake (OpenHands/coding-agent niche), not a mass-market chat model.

Capability

3 high confidence
3.0

Best open coding model in its release window at a small 24B size, but below 2026 frontier coders (Qwen3.6 27B 77.2, GLM-5.1, Kimi K2.6 80%+). Strong-for-size, mid-tier overall.

Unchanged since 2026-06-24 (last edited, not re-checked)