Overnight, OpenAI stitched together ChatGPT, Codex, and Work

Zhishi Author | Chen Junda Editor | Xinyuan Zhixi reported on July 10 that early this morning, OpenAI conducted an online live broadcast, officially launching the new generation GPT-5.6 model family to global users, and simultaneously released three major product updates of the Agent product ChatGPT Work, the new ChatGPT desktop application, and Hosted Sites (hosted websites). The GPT-5.6 series includes the flagship model GPT-5.6 Sol, the GPT-5.6 Terra that focuses on daily tasks, and the fast model GPT-5.6 Luna. Among them, Sol will be gradually opened to all paying users in the next 24 hours, while Terra and Luna will be opened to free users. On the third-party evaluation platform Artificial Analysis, GPT-5.6 Sol (max reasoning strength) scored 59 points, only 1 point behind Claude Fable 5. But its average cost per task is only US$1.04, while Fable 5 is US$2.75, the former is about 37% of the latter. On the programming list of Artificial Analysis, GPT-5.6 Sol scored 80 points, 2.8 points higher than Claude Fable 5. Judging from the benchmark test published by OpenAI, in some software engineering and difficult mathematical tasks, Claude Mythos 5 and Fable 5 still have the advantage; but in scenarios that are closer to daily applications such as Agent, computer operation and network search, GPT-5.6 Sol has achieved equal or even overtaken. ▲Comparison of key benchmarks between GPT and Claude models (data source is OpenAI blog, Zhixixi uses AI-assisted drawing) However, for OpenAI, today’s real major update is ChatGPT Work. Its positioning is similar to Claude Cowork or Kimi Work, which can help users complete information collection.