Mistral AIDevstral 2
Devstral 2 is a Mistral AI open-weights model, released 9 Dec 2025. Its best configuration ranks #600 of 785 on the BenchLeader Index at 42.3 ±4.4, in the lower half. The ± is the point: 301 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in agents & tools (42) and lowest in coding (36). At $0.800 per million tokens blended it is mid-priced. Output speed of 80 tokens per second puts it slower than most, with a first token in 0.7 s. Last measured 23 Oct 2026.
- Blended price
- $0.800/M
- $0.400 in · $2.00 out
- Output speed
- 80 tok/s
- OpenRouter traffic, 7-day median; not yet measured by Artificial Analysis
- First answer
- 0.72 s
- Context
- 262k
- Full answer
- –
- Cost per run
- –
- Released
- 9 Dec 2025
- Overall index42
- Reasoning41
- Coding36
- Agents & tools42
- Knowledge40
- Instruction following41
- Long context40
- Composite39
How its index has moved
9 Sept 2026 to 21 Oct 2026The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.
Benchmark results
See all LLM benchmarks. One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.
Reasoning
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| GPQA Diamond (AA)not in index | 59.4%#379 | Artificial Analysis | |
| Humanity's Last Exam (AA)not in index | 3.6%#586 | Artificial Analysis | |
| CritPt | 0.0%#315 | Artificial Analysis |
Coding
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Arena WebDev | 1197#134 | Arena (LMArena) | |
| SciCode (AA)not in index | 32.8%#204 | Artificial Analysis | |
| LiveCodeBench | 51.8%#110 | Vals AI | |
| SWE-bench (Vals)not in index | 62.8%#72 | Vals AI | |
| SWE-bench Verified (bash only) | 53.8%#31 | SWE-bench | |
| SWE-bench Verified (any scaffold)not in index | 53.8%#34 | SWE-bench | |
| Terminal-Bench 4.0 (AA)not in index | 0.0%#152 | Artificial Analysis | |
| Terminal-Bench 2.1 (AA)not in index | 30.3%#153 | Artificial Analysis |
Agents & tools
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| Terminal-Bench Hard | 18.9%#166 | Artificial Analysis | |
| τ²-Bench Telecom (AA)not in index | 24.9%#296 | Artificial Analysis | |
| τ³-Banking (AA) | 10.5%#141 | Artificial Analysis | |
| AutomationBenchnot in index | 3.1%#174 | Artificial Analysis | |
| GDP.pdf (AA)not in index | 2.4%#188 | Artificial Analysis |
Knowledge
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-Omniscience | -46.7#364 | Artificial Analysis | |
| AA-Omniscience: accuracynot in index | 20.8%#292 | Artificial Analysis | |
| AA-Omniscience: non-hallucinationnot in index | 14.7%#343 | Artificial Analysis |
Instruction following
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| IFBench | 38.1%#282 | Artificial Analysis |
Long context
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA-LCR | 32.3%#384 | Artificial Analysis |
Composite
| Benchmark | not stated | Source | Trend |
|---|---|---|---|
| AA Intelligence Index v4.3.2 | 8.6#432 | Artificial Analysis |
Where to run it
Every provider serving this model through OpenRouter, with throughput and first-token latency measured on live traffic over the last 30 minutes and each provider’s own price. Purple marks the best in each column.
| Provider | Speed | First token | Input $/M | Output $/M | Context | Quantisation |
|---|---|---|---|---|---|---|
| Mistral | 87 tok/s | 0.68 s | $0.400 | $2.00 | 262k | – |
What a task costs
Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.
| Workload | Tokens in / out | Cost | With caching | Time |
|---|---|---|---|---|
| Chat reply | 400 / 300 | $0.0008 | – | 4.5 s |
| Summarize a 30-page report | 12,000 / 600 | $0.0060 | – | 8.3 s |
| Code edit | 6,000 / 1,500 | $0.0054 | – | 19.6 s |
| Agentic coding session | 60,000 / 4,000 | $0.032 | – | 51.0 s |
| Structured extraction | 2,000 / 200 | $0.0012 | – | 3.2 s |
More from Mistral AI
See also
Questions people ask
- What are Devstral 2's benchmark scores?
- As of 25 Oct 2026, Devstral 2 scores 42.3 ±4.4 on the BenchLeader Index, strongest in agents & tools (42) and weakest in coding (36). That rests on 10 independent benchmark results, and every score on this page links to its publisher.
- How does Devstral 2 rank?
- Devstral 2 is #600 of 785 ranked configurations on 25 Oct 2026. 301 other configurations score within its margin, so the exact position matters less than the band. The leader is Claude Opus 5.5 (max effort) at 72.6.
- How much does Devstral 2 cost per 1M tokens?
- As of 25 Oct 2026, Devstral 2 costs $0.400 per 1M input tokens and $2.00 per 1M output tokens, $0.800 blended (three input tokens per output token).
- When was Devstral 2 released?
- Mistral AI released Devstral 2 on 9 Dec 2025.
Data as of 25 Oct 2026. Compare with another model.
Cite as: BenchLeader, “Devstral 2: benchmarks, pricing, speed and rank”, data as of 25 Oct 2026, https://www.benchleader.com/models/devstral-2.