Mistral: Codestral 2508 API Pricing
Mistral: Codestral 2508 costs $0.300 per 1M input tokens and $0.900 per 1M output tokens (cache reads: $0.030). Context window: 256K. Prices verified 2026-10-21 · source
Alternatives to consider
Cheapest mistralai model — 93% less per month
Rebuilt daily from our snapshot archive — the chart appears once a model has 5+ snapshots of history.
From our own daily snapshot archive — independent of provider announcements.
Token economics
| Token type | Price | vs base input |
|---|---|---|
| Standard input | $0.300 | — |
| Cache read | $0.030 | −90% |
| Output | $0.900 | — |
Prompt caching stores repeated context server-side: reads cost far less, but writes carry a one-time premium.
Training data ends here — newer events require retrieval or web search.
What would Mistral: Codestral 2508 cost you per month?
| Workload | Requests/mo | Avg tokens (in/out) | Est. monthly cost |
|---|---|---|---|
| Light chatbot | 10,000 | 800 / 400 | $6.00 |
| Coding agent | 5,000 | 8,000 / 2,000 | $21.00 |
| RAG pipeline | 30,000 | 4,000 / 500 | $49.50 |
| Doc summarizer | 2,000 | 20,000 / 1,000 | $13.80 |
Estimates use list prices without prompt caching. Caching can cut input cost by up to 90% on repeated context.
About Mistral: Codestral 2508
Mistral's cutting-edge language model for coding released end of July 2025. Codestral specializes in low-latency, high-frequency tasks such as fill-in-the-middle (FIM), code correction and test generation. [Blog Post](https://mistral.ai/news/codestral-25-08)
Compare Mistral: Codestral 2508 with
- Codestral 2508 vs OpenAI: GPT-5.6 Terra Pro
- Codestral 2508 vs Anthropic: Claude Fable 5
- Codestral 2508 vs Anthropic: Claude Opus 5
- Codestral 2508 vs OpenAI: GPT-5.6 Sol
Popular head-to-heads — see all on the comparisons index.
Frequently asked questions
How much does Mistral: Codestral 2508 cost?
Mistral: Codestral 2508 costs $0.300 per 1M input tokens and $0.900 per 1M output tokens; cached input tokens cost $0.030 per 1M. Verified 2026-10-21 — provider source.
What is the context window?
256K tokens, with up to 205K output tokens per response. That's roughly 320 pages of text in a single request.
Does it support prompt caching?
Yes — cached input is $0.030/1M, about 90% cheaper than standard input. Agents and RAG workloads with repeated context benefit most.
What's a cheaper alternative?
ScaleDown: Summarize saves about 94% on a typical workload with comparable quality.
Official resources
- Mistral: Codestral 2508 on OpenRouter — live endpoint, pricing source of record.
- Artificial Analysis leaderboard → — independent intelligence and performance rankings.
- Mistral models → — first-party model specifications and limits.
- Shred your current bill — paste a CSV and see if switching saves money.