Changelog

Product changes

• Response caching — Response caching is now available. Cache identical LLM requests to save cost and latency in agent workflows and testing.
• Presets: cache configuration UI — Presets now include a cache configuration section to control response caching behavior.
• Web search SDK fallback — Anthropic web_search tools are automatically converted to openrouter:web_search for SDK compatibility.
• DeepSeek V4 thinking support — DeepSeek V4 models now support the thinking parameter.
• Text-to-speech pricing and format fixes — Improved text-to-speech pricing display and fixed MP3 format rejection for Gemini text-to-speech. Text-to-speech docs
• Privacy settings banner — A banner now appears on the Models page when all model providers are hidden via privacy settings.