Give a model the screenshot of a broken page, the component behind it, and a test that defines the fix. That is a useful first trial for Mistral Large 4, now available through Mistral's API in public preview.
The October 6 release combines multimodal input, a million-token context window, function calling, and structured output. Those capabilities let a harness put visual evidence and code into one workflow. They do not tell us whether the resulting patch passes. Model documentation.
There is an availability detail worth getting right: API access is available; the open weights are forthcoming. The official changelog separates those milestones. If your requirement is deployment on your own hardware, the preview endpoint does not satisfy it yet.
Standard uncached input/output rates are $1.36 / $4.18 per million tokens. The launch offer halves them to $0.68 / $2.09 for two weeks. Our calculators interpret that as October 6–19 inclusive, with standard rates returning October 20. That end date is our calendar interpretation of the announcement, not a separately published cutoff. Official pricing.
Try the same task against your current model. Record retries, total cost, and what needed correcting. The model tools now include Large 4, while its developer benchmark results remain untested. A low token bill is interesting; a verified result is the useful part.