Mistral touts Large 4 coding gains on third-party benchmark

Mistral AI posts thanks to two outside testers, @ThibaultJaigu and @RequestyAI, after three weeks of testing its Large model update.

Mistral says the testing was a training exercise for the model. In a quoted post, Jaigu says the Mistral Large 4 update is much better at coding and ranks as the second-best open-source model on his coding benchmark. The benchmark's name, scores and test method are not given in the reports, and Mistral does not state its own figures.