v0.5.8 v0.28.0
MiniMax-AI/MiniMax-01Aug 24, 2026 → Aug 26, 20262 days between releases
What's New
Major version jump to VLLM engine with breaking changes and performance optimizations
Breaking Changes
- Bitsandbytes support migrated to an out-of-tree plugin
- Transformers bumped to 5.15.0
- Deprecated `calculate_kv_scales` removed
- `override_attention_dtype` removed
New Features
- Kimi-K3 optimizations including Decode Context Parallel (DCP) and fused kernels
- DeepSeek V4 sparse MLA support
- Model Runner V2 with E/P/D disaggregation and weight offloading
- Rust frontend with gRPC support and standalone renderer
- New model support: Muse Glimmer, Dots3 NOTE, Interns2mobius
Improvements
- Tiered KV cache offloading support
- Fused FlashKDA kernels for sequence parallelism
- AMD Quark NVFP4 support
- ROCm enablement for Kimi-K3
Upgrade Notes
Migrate Bitsandbytes to an out-of-tree plugin and update Transformers to 5.15.0. Deprecated functions `calculate_kv_scales` and `override_attention_dtype` have been removed. Expect significant architectural changes with the VLLM engine upgrade.
The Two Releases
Release v0.5.8
Aug 24, 2026· 2 months ago
Focuses on CI improvements and terminal rendering accuracy, specifically addressing UTF-8 width counting for Thai characters and mouse tracking.
Full release notesv0.28.0
Aug 26, 2026· 2 months ago
A major performance and feature release for the VLLM engine, featuring Kimi-K3 optimizations, DeepSeek V4 support, and Model Runner V2 maturity.
Full release notes