All times Eastern (ET)Provider pages checked daily at

Devstral 2

API ID devstral-2512

Devstral 2 was a Mistral model, released December 2025. It had a 256,000-token context window. It took PDF input and supported function calling and structured outputs.

Our frontier code agents model for solving software engineering tasks; excels at using tools to explore codebases, editing multiple files and power software engineering agents.

Mistral's description
Status
Retired
Released
December 2025
Shutdown
Jul 31, 2026

devstral-2512 stopped serving on July 31, 2026. Recommended replacement: mistral-medium-3-5.

How to migrate to Mistral Medium 3.5

Benchmarks

Scores as each source published them
or

Comparing with Devstral Medium 1.0, the previous model in this line. A score is compared only with the same benchmark from the same source, at the same setting where both have it. Full comparison: prices, limits, features

Arena

ArenaDevstral 2Devstral Medium 1.0ChangeRankDateSource
WebDev1196 ±211071 ±31+125137 of 142
1,182 votes
Oct 13, 2026source

Rated as devstral-2; Devstral Medium 1.0 as devstral-medium-2507. Ratings come from people voting between two anonymous answers. Data: Arena (arena.ai), CC BY 4.0.

On cloud platforms

All platform dates
PlatformPlatform IDStateShutdownTime leftReplacementSource
Amazon Bedrockmistral.devstral-2-123bLegacyMar 30, 2027166 days—source

Amazon Bedrock retires this model 242 days after Mistral's own API.

Read daily from each platform's own pages. The shutdown date and countdown on this page use Mistral's own date. Time left counts to 00:00 UTC on the shutdown date, which is the evening before in ET.

Aliases pointing here: devstral-latest (in place when tracking began (October 6, 2026)), devstral-medium-latest (in place when tracking began (October 6, 2026))

FeatureValueSinceSource
Structured outputsThe model can be held to a JSON schema you supply, so every reply parses without retries or cleanup.Supportedin place when tracking began (October 6, 2026)source
Function callingAlso called tool use. You describe functions; the model replies with a function name and arguments for your code to run, then uses the result.Supportedin place when tracking began (October 6, 2026)source
Batch APISend many requests as one job that runs in the background, usually finished within 24 hours, at a lower price than live requests.Supportedin place when tracking began (October 6, 2026)source
PDF inputPDF files can go in the prompt directly, without converting them to text first.Supportedin place when tracking began (October 6, 2026)source
Context windowThe most the model can work with in one request, in tokens: your prompt, conversation history and files. A token is about three-quarters of an English word.256,000 tokensin place when tracking began (October 6, 2026)source

Not published by Mistral: Prompt caching, Image input, Reasoning / thinking, Fine-tuning, Built-in web search, temperature parameter, top_p parameter, logprobs, Max output tokens.

Price history

USD per 1M tokens, standard tier
$0.00$0.50$1.00$1.50$2.00$2.50Oct 2026In $0.40Out $2.00Oct 5, 2026 to Oct 15, 2026: Input $0.40, Output $2.00
$0.00$0.50$1.00$1.50$2.00$2.50Oct 2026In $0.40Out $2.00Oct 5, 2026 to Oct 15, 2026: Input $0.40, Output $2.00

We started recording this model's prices on Oct 5, 2026. The history grows with each daily check.

Show as a table
FromTypePrice
Oct 5, 2026Input$0.40
Oct 5, 2026Output$2.00

Prices

USD per 1M tokens, every type and band, by tier
TypePrice $/1MIn effect
Standard
input$0.40
output$2.00

Changelog

Changes we detected on Mistral's pages

No changes recorded yet.

Last checked · Sources: Mistral models