v0.8.6
EricLBuehler/mistral.rsv0.8.6Jun 17, 2026by EricLBuehler
AI Summary
A feature release introducing OpenAI-compatible agent platform features, prebuilt binaries for easier installation, Anthropic API support, and significant performance improvements across CUDA and Metal backends.
Key Highlights
- OpenAI-compatible agent platform with skills and files API
- Prebuilt binaries and one-line install scripts
- Anthropic API support
- New models: Gemma 4, DiffusionGemma, and Qwen 3.5
- CUDA performance improvements and Prometheus metrics endpoint
New Features
- Anthropic API support
- Prometheus /metrics endpoint
- Prebuilt binaries with download-first install
- Gemma 4 12B support
- DiffusionGemma and block-diffusion models
- Qwen 3.5 support
- Online calibration for K-quants
- Paged flash-attention kernels
- BF16 CUTLASS 2.x MoE kernels
Full Release Notes
## What's Changed * feat(agentic): add improved logic for file outputs from shell, general execution by @EricLBuehler in https://github.com/EricLBuehler/mistral.rs/pull/2235 **Full Changelog**: https://github.com/EricLBuehler/mistral.rs/compare/v0.8.5...v0.8.6