Models·Mistral

Devstral 2 2512: what it does and what it costs

mistralai/devstral-2512

Mistral’s open-weight text model for agentic coding.

$0.40Input, per million tokens
$2Output, per million tokens
262KContext window
210KLongest answer

262K tokens of context — about 196,608 words, on a scale that ends at 2M.

What it can do

  • Tool use
  • Open weights
  • Takes text
  • Takes file
  • Gives text

Released 9 Dec 2025 · checked 3 Oct 2026.

About it

Devstral 2 2512 is a Mistral open-weight model specializing in agentic coding. It takes text and files and returns text, with tool support. Its context window is 262,144 tokens, and it can produce up to 209,715 tokens in one answer. API prices are $0.4 per million input tokens and $2 per million output tokens.

Review

In a line

Devstral 2 2512 is Mistral’s open-weight model specializing in agentic coding. It reads text or files and produces text.

What it's good at

The model specializes in agentic coding: coding work where an AI model can use tools as part of a task. Tool support is included.

Price in plain terms

The API costs $0.4 per million input tokens and $2 per million output tokens. Tokens are small pieces of text the model reads or writes.

At those rates, reading a 500-word page costs about 0.026 cents. A typical turn with 1,000 input tokens and 500 output tokens costs about 0.14 cents; 1,000 such turns cost about $1.40.

Context and limits

It can take in up to 262,144 tokens at once, roughly the amount of material that fits in its context window. It can produce up to 209,715 tokens in one answer.

It handles text and files as input and returns text.

Ways to use it

Its weights are open, and its identifier is mistralai/devstral-2512.

Explore Mistral models, open-weight models, or models with tool calling.

Verdict

Consider it if you need a text model for coding work that can use tools, and its long context window and per-token API pricing suit your workload. Its open weights may also matter if you want a model whose weights are available.

It is a less suitable fit if you need image input or image output: the listed input types are text and files, and its output is text.

Verdict

Devstral 2 2512 is worth considering for coding workflows that can make use of tools, especially when a large amount of material needs to fit in one request. Its API pricing is low for the example workloads listed, though actual costs depend on how many tokens a task uses. Open weights are another reason to consider it. Choose a different model if you need image handling or a non-text response; this model is listed for text and file input and text output.

  • Open weights are available.
  • Supports tools for agentic coding.
  • Takes text and files as input and returns text.
  • 262,144-token context window and up to 209,715 output tokens.
  • API prices are $0.4 per million input tokens and $2 per million output tokens.
  • It specializes in agentic coding, not described for other specialties.
  • Its listed inputs are text and files; its output is text.
  • API charges vary with the number of input and output tokens used.
  • Pick it ifPeople building coding workflows that use tools; Tasks that need a large context window
  • Skip it ifYou need image input or image output; You want a listed specialty beyond agentic coding

Pricing

API use costs $0.4 per million input tokens and $2 per million output tokens. The example cost is about 0.14 cents for one typical turn, or about $1.40 for 1,000 turns; actual charges depend on token use.

Questions

What is Devstral 2 2512?

It is Mistral’s open-weight model specializing in agentic coding.

Can it use tools?

Yes. Tool support is listed for agentic coding.

What can I give it?

It accepts text and files. Its output is text.

How much context can it handle?

Its context window is 262,144 tokens, the amount of material it can take in at once.

How much does its API cost?

$0.4 per million input tokens and $2 per million output tokens. A sample turn costs about 0.14 cents; 1,000 such turns cost about $1.40.

Where can I use it?

Its model identifier is mistralai/devstral-2512.

On these boards

More from Mistral