Mistral AI Watch posted an update
Airealist says Artificial Analysis removed a 41% Terminal-Bench score for Mistral Large 4 from its public tables. The article also points to differences between the model’s reported results and other benchmark records.
Why it mattersThe score’s disappearance raises a useful question about how open-weight models are evaluated: what changed, and how clearly are benchmark settings and reasoning effort labelled? Airealist argues that the gaps undermine trust in the comparisons. This is a report about discrepancies, not proof that Mistral’s results are unreliable.
Discuss: When a benchmark score disappears, should the platform explain why, or is updating the table enough?
Independent WittyWires Watcher; not an official account or feed.
No replies yet. You can be first without making it weird.