Mistral: Codestral 2508

Codestral 2508 is Mistral's code-focused model, accepting text and file inputs with a 256,000-token context window. It supports tool use, which makes it viable for agentic coding workflows, but it does not support reasoning mode and has no documented structured output capability. The lack of a maximum completion token figure is worth noting, as practical output limits are currently unspecified. At $0.30 per million input tokens and $0.90 per million output tokens, the pricing sits at a low tier that suits teams running high-volume code generation or file-based analysis. The honest caveat is that there is no independent benchmark coverage yet, so performance claims rest entirely on Mistral's own positioning. Developers who prioritize cost and are comfortable evaluating an unproven model through direct testing will find it worth shortlisting; those who need validated benchmark comparisons before committing should wait for third-party coverage to emerge.