I evaluated Codestral in Mistral Studio through a Rider and Payments-focused engineering agent, using TypeScript coding and code-review prompts. In one task, I specified that a display-name function should trim leading and trailing whitespace, collapse repeated whitespace, preserve capitalization and Unicode characters, and return an empty string for whitespace-only input. Codestral generated the function and Jest tests for the requested cases, using standard JavaScript string methods and a regular expression. The implementation was concise and straightforward to compare against each requirement. I also used follow-up prompts to examine payment-retry logic, idempotency, timeout handling, and possible duplicate processing in a separate illustrative code example. These sessions let me assess both its initial code generation and how it responds when asked to reconsider an earlier solution. Review collected by and hosted on G2.com.

The main limitation I observed was consistency across successive revisions. In the payment-processing example, Codestral identified risks involving duplicate charges, idempotency, and timeout behavior, but some proposed corrections still had retry-termination issues or inconsistencies between the implementation and the suggested tests. I had to inspect the revised code against the original requirements instead of relying on the explanation alone. In the display-name task, the generated implementation met the main requirements, although some of its additional commentary about whitespace edge cases was inaccurate. I would like follow-up responses to preserve earlier requirements, correct the identified defects completely, and keep the tests aligned with the final implementation. Review collected by and hosted on G2.com.