# Devstral 2
Devstral 2 was [[Mistral AI]]'s second-generation agentic coding model, the successor to the original [[Devstral]] (Devstral Small 2505). It served as the default driver behind Mistral's coding agent in the [[Mistral Vibe]] product until April 2026, when it was replaced by the merged flagship [[Mistral Medium 3.5]].
## Position in the line
- **[[Devstral]]** (May 2025) — first generation; Devstral-Small-2505; lightweight (single RTX 4090 / 32GB Mac); SWE-Bench Verified ~46.8%
- **Devstral 2** — second generation; default for Mistral Vibe coding agents until early 2026
- **[[Mistral Medium 3.5]]** (Apr 2026) — replaces Devstral 2 in Vibe; SWE-Bench Verified 77.6% per Mistral
## Why it matters in the version chain
- Marks the moment Mistral's coding line jumped from a small specialist model to a general flagship model
- Replacement signals the consolidation strategy — Mistral is folding coding into one merged model rather than maintaining a separate Devstral line
- Reference point for "what was the open coding-model frontier on Mistral's side before the merge"
## Caveats
- Public detail on Devstral 2 (architecture, parameter count, benchmark suite) is sparse compared to the original Devstral release post; most facts come from the Medium 3.5 launch material that frames Devstral 2 as the predecessor
- Whether Devstral 2 remains available as a distinct API endpoint after the Medium 3.5 cutover is unclear — check current [[La Plateforme]] docs
## References
- Mistral Medium 3.5 launch post (positions Devstral 2 as the model being replaced): https://mistral.ai/news/vibe-remote-agents-mistral-medium-3-5
- Original Devstral announcement (for the line context): https://mistral.ai/news/devstral
## Related
- [[Mistral AI]]
- [[Devstral]]
- [[Mistral Medium 3.5]]
- [[Mistral Vibe]]
- [[Codestral]]
- [[Mistral Small 4]]
- [[Large Language Models (LLMs)]]