Devstral 2
Devstral 2 is behind the leaders in factuality and coding. Too few results yet to rate reasoning, agentic tasks, safety, long context, math, multimodal tasks, multilingual tasks, or instruction following.
Capability profile
Bars show the model's median result as a share of the leading model's, per capability. Select a row to see its results.
Too few results yet to rate Reasoning, Agentic, Safety, Long Context, Math, Multimodal, Multilingual or Instruction Following.
Price
$0.40input$2.00outputper million tokens
From mistralai · All prices
Evidence
24results on22benchmarks
- 3 independently verified
- 18 aggregator
- 3 vendor-reported
From 4 sources · latest Oct 8, 2026 · How verification works
API features
Tool callingStructured outputs
As listed by OpenRouter
Devstral 2 benchmark results
24 results on 22 benchmarks, grouped by capability. Each bar is the result as a share of the capability leader's; every result links to its source.
38.7% behind the leader2 of 4 ranked benchmarks measured
- AA-Omniscience · Accuracy20.82Oct 8, 2026omniscienceAccuracy
- AA-Omniscience · Non-hallucination14.73Oct 8, 2026omniscienceNonHallucination
44.1% behind the leader6 of 10 ranked benchmarks measured
- 72.20Jun 15, 2026
- 61.30Jun 15, 2026
- SciCode32.75Oct 8, 2026aa_scicode
- Terminal-Bench 2.130.34Oct 8, 2026terminalbenchV21
- Terminal-Bench Hard18.94Oct 8, 2026aa_terminalbench_hard
- 1193.87May 22, 2026
Show 2 more coding resultsHide 2 coding results
- 1199.13Jun 19, 2026
- 53.80Sep 8, 2026
0 of 6 ranked benchmarks measured
- 0.00Oct 8, 2026
- GPQA Diamond59.39Oct 8, 2026gpqa
- Humanity's Last Exam3.57Oct 8, 2026aa_hle
0 of 7 ranked benchmarks measured
- 2.37Oct 8, 2026
- Terminal-Bench 4.00.00Oct 8, 2026
- τ-Bench V3 · Banking10.52Oct 8, 2026tauBanking
0 of 3 ranked benchmarks measured
- 32.33Oct 8, 2026
0 of 3 ranked benchmarks measured
- IFBench38.10Oct 8, 2026aa_ifbench
More results
Benchmarks outside the capability baskets. They are not ranked against a leader.
- τ²-Bench Telecom (AA run)24.85Oct 8, 2026aa_tau2
- AA Intelligence8.62Oct 8, 2026aa_intelligence_index
- -46.70Oct 8, 2026
- Artificial Analysis Coding Index31.26Sep 9, 2026aa_coding_index
- 4.90Sep 9, 2026
- Terminal-Bench 2.032.60Jun 15, 2026Terminal Bench 2
Devstral 2: common questions
Who makes Devstral 2?
Devstral 2 is made by Mistral.
When was Devstral 2 released?
Devstral 2 was released on Dec 9, 2025, according to Mistral's own announcement.
What is Devstral 2 good at?
Devstral 2 is behind the leaders in factuality and coding. Too few results yet to rate reasoning, agentic tasks, safety, long context, math, multimodal tasks, multilingual tasks, or instruction following.
How much does Devstral 2 cost?
Devstral 2 costs $0.40 per million input tokens and $2.00 per million output tokens, according to mistralai. At a mix of three input tokens to one output token, it costs more than 52% of the 331 priced models we track.
How many benchmarks has Devstral 2 been tested on?
We track 24 results for Devstral 2 on 22 benchmarks from 4 sources, 3 of them independently verified. The latest was recorded on Oct 8, 2026.
Which API features does Devstral 2 support?
OpenRouter lists tool calling and structured outputs for Devstral 2.
About this record
Where Devstral 2's numbers come from, and every name it appears under.
- Tracked since
- May 2, 2026
- Newest source mention
- Sep 8, 2026
Where the results come from
Verification: 24 scores · 3 independently verified · 18 aggregator-attributed · 3 vendor-reported. How these tiers are assigned
From 4 sources on 4 sites. Artificial Analysis supplies 18 of them; the 3 independently verified results come from 2 sites. Bars are coloured by trust tier.
- artificialanalysis.ai18
- huggingface.co3
- datasets-server.huggingface.co2
- swebench.com1