February 27, 2026
Verified 72B model benchmarks showing 79× speedup over bitsandbytes.
Verified Qwen 72B benchmark: 6.5s cold start (79× faster than bitsandbytes)
llama.cpp GGUF comparison: ZSE 1.6× faster on 72B models
Updated website with H200 GPU benchmark results
New benchmark scripts for 70B+ models
February 25, 2026
Improved documentation and fixed PyPI classifier issues.
Fixed invalid PyPI classifier (removed CUDA language)
Updated README with verified 32B benchmarks
Added honest VRAM threshold notes
PyPI badge in README
February 24, 2026
Minor bug fixes and documentation updates.
Fixed package installation with optional GGUF support
Improved error messages for missing models
Support for custom model paths
February 23, 2026
First public release of ZSE with core functionality.
zQuantize: INT4/NF4 pre-quantization with .zse format
zServe: OpenAI-compatible API server
zInfer: CLI inference tool
GGUF import support
Streaming token generation
Multi-model management
Check out our roadmap and upcoming features on GitHub