A full pass against the providers' own pages. Joining: new reasoning and coding models from OpenAI, a smaller realtime voice model and a transcription model, a new Claude, two new Gemini tiers, a new music model and a new Grok. Two models were repriced, several end-of-life ones retired, and a shutdown date that would have switched three live models off in October turned out not to exist. DeepSeek's Flash tier, Seedance 2.5 video and the new OpenAI image model landed the same week.