OpenAI Brings GPT-6 to ChatGPT, but Only Inside Work and Codex

Instead of appearing in the familiar model picker on the right-hand side of the prompt bar, GPT-6 Astra, Sol and Luna are available in Work and Codex only, TechRadar reported. The article's author, a ChatGPT Plus subscriber, said that switching to the Work tab inside a Plus account exposes GPT-6 Luna High, GPT-6 Sol Light, GPT-6 Sol Medium, GPT-6 Astra Light and GPT-6 Astra Medium in the model picker.

OpenAI describes GPT-6 Astra as "the most intelligent and aligned model in the world," positioning it as the heaviest model in the family and intended for difficult jobs where maximum intelligence matters more than speed, according to the report. The company emphasizes Astra's ability to operate software, browse, conduct research and complete multistep professional workflows.

Alongside it, OpenAI released two smaller models, GPT-6 Sol and GPT-6 Luna, which the company says are more lightweight and less expensive to run. They are trained using similar methods to Astra and, in OpenAI's words cited by TechRadar, "build on the advances behind GPT-6 Astra," bringing much of its capability into faster and more affordable options.

The rollout is not entirely new. Astra has been available since earlier this month, and GPT-6 Pro, which is powered by Astra, is offered in regular Chat on some higher-tier plans, the report said. The obstacle for other users is plan tier rather than model availability: a Plus subscriber can reach Astra and the new Sol and Luna models only through Work and Codex.

TechRadar's account draws a working distinction between the two settings. In Chat, the author describes working alongside the model — asking something, reading the response, adding detail and steering the conversation. In Work, more of the process is delegated: a larger task with multiple steps can be handed over for the model to research, inspect files or use websites without step-by-step instruction.

Both modes use the same interface, with a prompt bar at the bottom of the screen, which the report identifies as one reason the split is confusing. Using Work for ordinary conversation is possible but slow, the author found, because Work is built around longer multi-step jobs rather than quick conversational replies.

Until this launch, the author said, it was possible to spend all of one's time in Chat on GPT-5.6 Sol and largely ignore Work. The placement of the new models changes that calculation, since the most capable GPT-6 variants now sit on the Work side of the divide. The report did not include comment from OpenAI.