Paid OpenRouter models, more reliable Pro providers, and Gemini back in the Pro provider list.
Added
- A Also show paid models, with pricing option in the OpenRouter connect panel. Licensed sites can browse OpenRouter’s full catalog and see each model’s price per million tokens.
- Live model discovery for Anthropic, Groq, Mistral, and DeepSeek, so each provider’s current model list is fetched instead of a fixed default that can go stale.
Changed
- Google Gemini is a Pro provider again, alongside Anthropic, Groq, Mistral, DeepSeek, and Ollama. No action is needed if Gemini is already connected and licensed.
- If your Pro license lapses while a paid OpenRouter model is saved, generation keeps working on a free model, and the OpenRouter card explains why. Nothing switches to a paid model automatically when Pro is activated or upgraded.
Fixed
- Several Gemini reliability issues: a blocked default model, a raw 429 quota error with no plain-language explanation, and a key test that could fail with “Empty response from provider.”
- Four more dead or deprecated model defaults across Groq, Mistral, DeepSeek, and GPT Image thumbnails, each replaced with the provider’s current recommendation.
- Ollama reliability: requests to localhost could hang, so they now use 127.0.0.1. The timeout is 300 seconds for slower hardware, and a timeout race that could cause a fatal error is resolved.
- The SEO minimum-score retry loop could discard a real generation failure during a regeneration. It now reports the actual error.
