Codestral
Codestral by Mistral AI: $0.3 input and $0.9 output per 1M tokens, 256K context. Call it through the odnoga LLM gateway.
Codestral is served by Mistral AI and called through odnoga with the same OpenAI-compatible request shape as every other model in the catalog. It supports tool and function calling. It can be forced to return structured JSON. Its 256K-token context window sets how much input you can send in one request.
Specification and price
| Vendor | Mistral AI |
|---|---|
| Model ID | codestral-latest |
| Context window | 256K tokens |
| Max output | 8K tokens |
| Capabilities | Tools, JSON mode, Streaming |
| Per 1M tokens | Vendor cost | Your price on Free+7% |
|---|---|---|
| Input | $0.3 | $0.323 |
| Output | $0.9 | $0.968 |
| Cached input | — | — |
Vendor cost is the list price per million tokens as recorded in the odnoga catalog; your price applies your plan margin with the same formula that bills every request — see pricing. Pricing
Call it through odnoga
const res = await fetch("https://api.odnoga.com/v1/chat/completions", {
method: "POST",
headers: {
Authorization: `Bearer ${process.env.ODNOGA_API_KEY}`,
"Content-Type": "application/json",
},
body: JSON.stringify({
model: "codestral-latest",
messages: [{ role: "user", content: "Hello" }],
}),
});
Same request shape for every vendor in the catalog — swap the model id and odnoga handles keys, routing, limits and cost accounting.
How to use Codestral
Derived from the odnoga catalog record for this model.
Best for
- High-volume, latency-sensitive calls: classification, extraction, routing, short rewrites and summaries.
- Agents and workflows that call your own functions, because the model supports tool calling.
- Machine-readable output you can write straight into a database, using enforced JSON.
- Long inputs: a 256K-token window fits whole contracts, codebases or transcripts in one request.
- User-facing chat where partial output should appear while the model is still writing.
Not the right pick when
- Anything that needs to read an image — this model takes text only.
- Anything where a wrong answer is costly without a human check — no model in the catalog removes that requirement.
Practical tips through odnoga
- 01Pin the model id in a managed prompt version, so a model swap is a version change you can compare and roll back, not an edit in application code.
- 02Compare it against 2–8 other models on the same frozen test cases in the evaluation laboratory before you make it the production default.
- 03For repeated identical deterministic calls, odnoga answer reuse returns the stored answer and bills no vendor tokens — turn it off for creative output.
- 04Ask for JSON through the response format rather than in the prompt text — the schema is enforced instead of suggested.
- 05Set a fallback model on the route so a vendor incident degrades quality instead of returning an error, and a per-tenant budget so one caller cannot spend the month.
What a month costs
1,000 calls a month, 10,000 input tokens and 2,000 output tokens each, at your Free price (vendor cost +7%):
| Input (10M tokens) | $3.23 |
|---|---|
| Output (2M tokens) | $1.94 |
| Your cost per month on Free | $5.16 |
Vendor list cost $4.80 + odnoga margin $0.361 (+7%). Cached input or answer reuse lowers it; odnoga records both numbers per request.
Codestral compared
| Model | Context | Input / 1M | Output / 1M | Capabilities |
|---|---|---|---|---|
| Codestral | 256K | $0.3 | $0.9 | Tools, JSON mode, Streaming |
| Ministral 3B | 131K | $0.1 | $0.1 | Tools, JSON mode, Streaming |
| Ministral 8B | 262K | $0.15 | $0.15 | Tools, JSON mode, Streaming |
Questions
- How much does Codestral cost per 1M tokens?
- Mistral AI lists $0.3 / $0.9 per 1M input / output tokens in the odnoga catalog. Through odnoga you pay that vendor price plus your plan margin, and every request is recorded with both numbers.
- What is Codestral best for?
- High-volume, latency-sensitive calls: classification, extraction, routing, short rewrites and summaries. Agents and workflows that call your own functions, because the model supports tool calling. Machine-readable output you can write straight into a database, using enforced JSON.
- Can I switch to Codestral without changing my code?
- Yes. odnoga exposes one OpenAI-compatible endpoint, so switching means sending "codestral-latest" as the model id — or changing it in the managed prompt version, with no application deploy.
- How large is the Codestral context window?
- 256K tokens of input, with up to 8K tokens of output per response.
All models · Pricing · Docs · Compare models in the evaluation lab