Devstral 2
Mistral AI’s 123B dense coding model, released 2025-12-09; status: deprecated.
At a glance
- Context window
- 256K tokens
- Max output
- —
- Historical text input / output · per 1M tokens
- $0.400 / $2.00
- Knowledge cutoff
- —
A place to remember
Retired, not forgotten.
- Released
- 9 Dec 2025
- Endpoint retired
- 31 Jul 2026
- Time in service
- 234 days
Retired by provider
“Ranked on 27 leaderboards. The endpoint is gone. The weights are not.”
The endpoint retired. The open weights can still be run independently; this is not the death of the weights.
27 leaderboards · 0 top-three finishes · Best rank #31. The full model record remains below.
Benchmark snapshot
Best result in each categories
CodingTerminal Bench Hard
18.9Normal (With Tools) · #96 of 157
Source: Artificial Analysis (artificialanalysis.ai)Computer UsePinchBench v2
69.4Reported best (effort unspecified) · #31 of 51
Professionalτ³-Banking
10.5Normal (With Tools) · #91 of 120
Source: Artificial Analysis (artificialanalysis.ai)
Benchmark details
3 results · Best published score per benchmark, grouped by category
| Benchmark · setting | Score | Rank |
|---|---|---|
| Coding | ||
| Terminal Bench Hard Normal (With Tools) · Agentic Development Source: Artificial Analysis (artificialanalysis.ai) | 18.9 | 96 / 157 |
| Computer Use | ||
| PinchBench v2 Reported best (effort unspecified) · Tool Orchestration | 69.4 | 31 / 51 |
| Professional | ||
| τ³-Banking Normal (With Tools) · Service Workflows Source: Artificial Analysis (artificialanalysis.ai) | 10.5 | 91 / 120 |
Rank is this model's position among every model with a published result on that leaderboard; the bar shows how far up the field that is. Orange marks first place on leaderboards with at least 5 models. Unranked leaderboards publish a composite or direction-less score.
Back to benchmark snapshot ↑Historical API pricing
Listed text rates, USD per million tokens · effective 2026-11-14
- Input
- $0.40
- Output
- $2.00
Blended price4:1 input:output · same basis as the price ranking
$0.72/ 1M tokens
Specifications
- Provider
- MistralAI
- Model type
- Coding model
- Released
- December 9, 2025
- Endpoint retired
- 2026-07-31 Source
- Total parameters
- 123B
- Architecture
- Dense
- Context window
- 256K tokens
- Input
- Text
- Output
- Text
- Weights
- Open weights
- Licence
- Modified MIT
- Commercial use
- Commercial use allowed
Capabilities
Model analysis
Devstral 2 is Mistral AI’s 123B dense coding model: released 2025-12-09 per the official Mistral model card; API ID devstral-2512; parameters: 123B; context window 256k; license: Modified MIT.
Current status: deprecated
The status comes from the lifecycle shown in Mistral Docs. Note: 123B dense model. This entry is based on the official model page only; independent benchmarks and pricing have not been added.
Source: Mistral official page. Retrieved 2026-10-06.
Last reviewed 2026-10-06
Devstral 2 vs competitors
Comparable results only · record is Devstral 2's wins–losses
| Benchmark | Devstral 2This model | vs DeepSeek-V4-ProRecord 1–2 | vs GPT-5.6 SolRecord 0–3 | vs GPT-5.6 TerraRecord 0–3 | vs GLM-5.2Record 0–3 |
|---|---|---|---|---|---|
| Terminal Bench Hard | 18.9 | 46.2 | 65.9 | 62.9 | 50.8 |
| τ³-Banking | 10.5 | 30.1 | 46.9 | 40.2 | 37.1 |
| PinchBench v2 | 69.4 | 61.1 | 84.2 | 75.9 | 87.0 |
Primary sources
- Devstral 2 technical reportdocs.mistral.ai
On the stone
Memories and epitaphs, kept separately from model reviews.
Epitaphs from visitors
Nobody has written on this stone yet. The first line is yours.
Community discussion
Share the API, setup, and tradeoffs you actually tested. Rate the model once; comment as often as you like.
Used this model? Sign in with Google to rate it and share what worked, what did not, and which API or setup you tested.
No community comments yet.
