March 2026 #

  • model released Magistral Medium 1.2 — state-of-the-art reasoning with magistral-medium-2506
  • api updated Added thinking_tokens field to chat completions response for reasoning traces
  • new Batch inference endpoint now supports up to 50,000 requests per job
  • docs Added agent orchestration cookbook with tool-calling examples

February 2026 #

  • model released Ministral 3B and 8B now available via API — ministral-3b-2410, ministral-8b-2410
  • api updated Fine-tuning API now supports validation_file parameter for automatic eval during training
  • deprecated mistral-tiny scheduled for removal on June 1, 2026 — migrate to ministral-3b-2410

January 2026 #

  • security Rotated signing keys for JWT tokens — no action required, tokens are renewed on next authentication
  • new Function calling now supports parallel tool execution — multiple tools can be called simultaneously in a single response