v3.5.4 v1.12.5
cheaper-inference/OmniRouteApr 7, 2026 → Apr 8, 20261 day between releases
What's New
Major version jump introduces Fish Speech S2 Pro engine with GPU acceleration
New Features
- Fish Speech S2 Pro engine integration
- Auto-download from HuggingFace
- Support for [tag] syntax
- Persistent caching in models/.cache
- Windows compilation patches
Improvements
- Triton/Inductor GPU kernel compilation
- Persistent caching for faster generations
- Windows compilation compatibility patches
- Informative first-run cache warnings
Upgrade Notes
The major version jump (3.5.4 to 1.12.5) suggests significant architectural changes. Ensure configuration files are compatible with the new TTS engine settings, as persistent caching is now enabled in the models directory.
The Two Releases
v3.5.4
Apr 7, 2026· 7 months ago
Enhances token tracking, adds legacy JSON config import/export, and improves Russian localization. Fixes Anthropic streaming input undercount and OAuth expiry display.
Full release notesVersion 1.12.5 - Fish Speech with Compilation Speed-Up
Apr 8, 2026· 7 months ago
This release introduces the Fish Speech S2 Pro engine, significantly enhancing multi-language support and control capabilities. It also focuses on performance improvements through GPU kernel compilation and persistent caching to accelerate generation times.
Full release notes