Overview
- OpenAI on July 10 rolled out the GPT‑5.6 family (Sol, Terra, Luna) across ChatGPT, Codex and its API after the U.S. Department of Commerce completed supplemental testing and approved broader access.
- Alongside GPT‑5.6, OpenAI launched ChatGPT Work for long multi‑step workplace tasks, introduced GPT‑Live voice models for full‑duplex conversation, and added gpt‑realtime‑2.1 models to its API to cut p95 latency by at least 25%.
- SpaceXAI released Grok 4.5, jointly trained with Cursor, and said the model halves token use versus peers, runs faster at high token throughput, and is priced cheaper for developers; Meta opened its Muse Spark 1.1 API to developers the same week.
- Vendors are trading on measurable edges: OpenAI says Sol is about 54% more token‑efficient in key tests, SpaceXAI and Meta announced lower per‑token pricing and performance claims, and Microsoft confirmed wide Copilot integration of the new models.
- The releases follow weeks of government‑led reviews that set phased access rules and export controls; the near‑term effects include faster embedding of advanced models into office tools, shifting costs for app builders, and continued regional rollout limits such as staged EU availability.