A weekly AI news roundup covering an exceptionally busy week dominated by OpenAI releases: GPT-5.5 and GPT-5.5 Pro (new SOTA LLM with 40% fewer tokens vs predecessor), GPT-Image-v2 (biggest Arena Elo jump ever, capable of functional QR codes and 360 images), and Codex updates with Computer Use and Chronicle. Anthropic launched Claude Design, a research preview UI tool that rattled Figma's stock price. Open source highlights include Kimi K2.6 (1T parameter MoE, 15x cheaper than Opus) and Qwen 3.6 27B (runnable on 18GB RAM, matching Sonnet 4.5 on coding benchmarks). Also covered: Google Gemini Deep Research Max, a ChatGPT Agents 'Hermes' leak, OpenAI's Privacy Filter open-source model, and a rumored SpaceX/Cursor $60B deal.
Table of contents
OpenAI’s GPT 5.5 is here - SOTA AI intelligence you can actually use ( Release Blog )GPT Image v2 - SOTA thinking image model, finally beating Nano Banana ( Blog , Live )Claude Design - Anthropic’s figma killer? ( try it , deep dive )Open Source AI6 Impressions