AI Round-up: June 25, 2026

OpenAI's Jalapeño chip led the snapshot. OpenAI and Broadcom unveiled a custom inference chip for ChatGPT, Codex, the API and future agentic products, with OpenAI's own post amplified across X and Reuters and Bloomberg framing the move as a push for faster, cheaper model serving. OpenAI also pushed a more conversational GPT-5.5 Instant, while Google DeepMind and Together AI kept the agent-infrastructure thread active through agent-economy discussion, GLM-5.2 comparisons and production-inference posts.

The counterweight was model governance, medical reanalysis and frontier-lab pressure. Reuters and Bloomberg tracked Anthropic's claims that Alibaba illicitly accessed Claude capabilities, China's 360 claimed Mythos-matching tools, and Bloomberg reported more Google AI staff moving to Anthropic. Microsoft published Talos for continuous genomic reanalysis across nearly 5,000 unsolved cases, while an OpenAI-linked immunology thread showed GPT-5 Pro revisiting a three-year-old experiment. YouTube and Reddit stayed centred on Claude/Fable fallout, Copilot and Codex workflow issues, Claude Design, Mistral OCR and AI-search noise.