Mycelis v0.3: smart routing, virtual models, and new GPU types

With Mycelis v0.3 we are shipping three core improvements that make operations noticeably easier.

1) Virtual models

A virtual model bundles multiple deployments behind one stable API slug.

• no hard-coded model names in clients
• quick backend swaps
• consistent API behavior across teams

2) Smart routing

Routing rules now decide per request which deployment fits best.

1. Lowest-cost model for standard requests
2. High-quality model for critical prompts
3. Fallback during traffic spikes or failures

3) New GPU types

Additional GPU types for different load profiles help balance cost and latency.

Tip: In production, combine a low-cost default deployment with a high-quality deployment.