GPT-6 Sol leaks, screenshots and release rumors

Someone shared a Codex multi-agent config that calls openai-codex/gpt-6-sol and gpt-6-luna for different steps, each with a thinking level.

The author ran the same short-video editing prompt through a model and asks commenters to guess whether Claude Opus 5.5 or GPT-6 Sol produced the clip.

Front-end developer Wes Bos posted that he is testing GPT-6 Sol with 3D rendering, an early hands-on signal with no results shared yet.

A Raycast employee shared that Sol and Luna are now available, claiming Sol doubles accuracy at half the cost while Luna is positioned as a highly efficient workhorse.

A post shares OpenRouter TPS figures: Opus 5.5 Fast exceeds 150 tok/s, GPT-6 Sol does 52 standard and 55 fast, and Astra 25, questioning the value of fast-mode surcharges.

A video post jokes that a typhoon, the Opus 5.5 release and OpenAI's GPT-6 Sol/Luna all landed on the final day of a holiday, wishing AI could handle all the work ahead.

Higgsfield released a video comparing Claude Opus 5.5 and GPT-6 Sol on the same 3D game development task.

A joking post notes that Grok 4.7, Claude Opus 5.5 and GPT-6 Sol all launched in the same week, contrasting with last week's calls to slow down AI.

A post says the GPT-6 Sol and Luna release comes with a banked reset and higher usage limits, and the author admits being more interested in those than in the models.

A Turkish-language post lists Grok 4.7, Opus 5.5 and GPT-6 Sol/Luna launching on the same day, and jokes that Chinese labs will rush out distilled models next week.

A satirical post mocks US AI labs for releasing Grok 4.7, Opus 5.5 and GPT-6 Sol just days after saying the frontier must be paced.

A Portuguese post claims GPT-6 Sol will be the cheapest frontier model, Luna cuts cost 60% versus the previous generation to replace open-weight models, and Grok 4.7 isn't worth upgrading to.