For coding agents, Devstral 2 costs 1.4x as much as Codestral per request ($0.0099 vs $0.0070 on our measured traffic), because Devstral 2 charges $2 per million output tokens against Codestral's $0.90.
Run current Claude, GPT, Gemini and open models in your agent
Or let the router send each step to the cheapest model that can do it, on providers we vet for price, uptime and speed. Plans from $19 a month.
| Cost at list price | Codestral | Devstral 2 |
|---|---|---|
| Prompt under 100K tokens50% of requests, 87% of input cached | $0.0043 | $0.0062 |
| Prompt of 100K to 272K39% of requests, 88% of input cached | $0.011 | $0.015 |
| Prompt over 272K tokens11% of requests, 86% of input cached | n/a | n/a |
| Average request | $0.0070 | $0.0099 |
| 100 requests | $0.70 | $0.99 |
| Median session16 requests; a session ends after 10 idle minutes | $0.084 | $0.12 |
| A month of regular usemedian of 20 users active on 5+ days | $8.91 | $12.53 |
Each logged request priced at the vendor’s list rates for its own token counts, including cache reads and any higher rate for long prompts. Average output: 698 tokens per request. Both models are priced on the requests up to 256K tokens that fit in both context windows; band shares above describe all requests.
They are built for different jobs: Codestral for autocomplete and quick edits, Devstral 2 for agent loops, so the cheaper one is not a drop-in swap. Both windows (256K and 262K) are smaller than about 12% of coding-agent requests. Devstral 2's price is Mistral's endpoint on OpenRouter, because Mistral's own price page does not list it.
Pick a winner in each dimension. You can change your vote any time. 4 votes cast so far.
Correctness and depth of what it produces
Tool calls, instruction-following, and multi-step tasks
Tokens per second and time-to-first-token
How much capability you get per dollar
Consistent results: fewer refusals, loops and format breaks
The agent matters as much as the model. Compare AI agents →
Current Claude, GPT, Gemini and open models run on Standard Compute