AI Daily Brief — 29 April 2026

Mistral AI ships a three-part release: Mistral Medium 3.5 (128B dense), Vibe remote coding agents over CLI, and Work Mode in Le Chat for multi-step autonomous tasks. The model lands to a muted reception — community benchmarks point out that Alibaba’s Qwen 3.6 27B hits 72.4% on SWE-Bench Verified at a quarter of Medium 3.5’s parameter count, under Apache 2.0.

Top stories

• Mistral Medium 3.5. A 128B dense model with strong reasoning + agentic features. Pricing: $1.50 per M input tokens, $7.50 per M output. via Mistral
• Mistral Vibe remote agents. Cloud-based coding agents launched from the Vibe CLI — can push pull requests to GitHub and run in parallel without an open terminal. via Mistral Docs
• Le Chat — Work Mode. Mistral’s consumer chat surface now handles multi-step autonomous tasks: email triage, research synthesis, cross-tool workflows. via Yahoo Tech
• Reception turns to comparisons. Coverage flags Qwen 3.6 (27B, Apache 2.0) hitting 72.4% on SWE-Bench Verified — at roughly a quarter of Medium 3.5’s parameter count — as the headline contrast. via Decrypt

By the numbers

• Mistral Medium 3.5 pricing: $1.50 / $7.50 per M tokens (input / output)
• Qwen 3.6 27B SWE-Bench Verified: 72.4% · Apache 2.0 · free to self-host

Compiled by AI Feed’s editor from publicly reported announcements on 29 April 2026.