Mistral AIOpen weightsAuto-detected

Devstral 2

Name: Devstral 2 benchmark results
Creator: BenchLeader

Best configuration ranks #442 of 610 on the BenchLeader Index at 43.1 ±6.4. Last measured 8 Sept 2026. Released 9 Dec 2025.

Blended price: –
Output speed: 126 tok/s
First answer: 2.30 s; first token 2.30 s
Context: 256k

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index43
Coding23
Agents & tools50
Knowledge42
Instruction following41
Long context42
Composite41

Benchmark results

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	default	Source	Trend
GPQA Diamond (AA)not in index	59.4%#360	Artificial Analysis
Humanity's Last Exam (AA)not in index	3.6%#499	Artificial Analysis

Coding

Benchmark	default	Source	Trend
LMArena WebDev	1194#113	LMArena
SciCode (AA)not in index	32.8%#140	Artificial Analysis

Agents & tools

Benchmark	default	Source	Trend
Terminal-Bench Hard	18.9%#161	Artificial Analysis
τ²-Bench Telecom (AA)not in index	24.9%#285	Artificial Analysis

Knowledge

Benchmark	default	Source	Trend
AA-Omniscience	-46.7#293	Artificial Analysis

Instruction following

Benchmark	default	Source	Trend
IFBench	38.1%#269	Artificial Analysis

Long context

Benchmark	default	Source	Trend
AA-LCR	32.3%#312	Artificial Analysis

Composite

Benchmark	default	Source	Trend
AA Intelligence Index	9.4#330	Artificial Analysis

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	–	–	4.7 s
Summarise a 30-page report	12,000 / 600	–	–	7.1 s
Code edit	6,000 / 1,500	–	–	14.2 s
Agentic coding session	60,000 / 4,000	–	–	34.1 s
Structured extraction	2,000 / 200	–	–	3.9 s