Mistral’s ‘Magistral’ Reasoning Model Tops GPT-5 on Maths but Stumbles on French Law
Mistral's new open-weight reasoning model beats GPT-5 on competition maths yet underperforms on the European legal tasks it was pitched to handle.
Mistral's new open-weight reasoning model beats GPT-5 on competition maths yet underperforms on the European legal tasks it was pitched to handle.
Meta's open-weight Llama 5 ships with built-in agentic tooling, reigniting the open-versus-closed debate and pressuring OpenAI's API pricing.
Mistral says its compact Edge models match last year's flagship quality while running fully offline, challenging the cloud-API economics that fund rivals.
Have a question, tip, or story idea? We read every message.