GPT-6 Sol leaks, screenshots and release rumors

The author used GPT-6 Sol with Combos CLI to build a 2-4 player online forest Monopoly game for his son, covering modeling, voice and deployment, with a playable link.

A user shares a Codex workflow for making 3D games where Astra plans the approach and multiple Sol subagents execute it, calling Sol cheap and effective.

A user says GPT-6 Sol xhigh has run for over five hours using only 1-2% of the usage limit, with quality still unjudged but far less token-hungry than Astra.

A poll for developers asks which of Claude Opus 5.5, GPT-6 Sol and Grok 3.7 they would pick as their main model.

After using Sol since launch, the poster says it feels more like Terra and is priced close to 5.6 Terra, suspecting OpenAI upgraded and renamed Terra and that a real Sol upgrade may come at Dev Day.

A subjective ranking places Opus 5.5 and Fable 5.1 above GPT-6 Sol, concluding Claude won this update.

A screenshot shows Opus 5.5 at 6% on the tough MazeBench 3D spatial reasoning benchmark, while GPT-6 Sol and Grok 4.7 both score 1%.

A post lists Grok 4.7, Claude Plus 5.5 and GPT-6 Sol/Luna as the three options and asks which one people would trust as their primary AI model.

A developer was wowed by Opus 5.5 in the morning, then sounded deflated after trying GPT-6 Sol in the afternoon, with full test details promised in a video tomorrow.

Testing four models on the same kanban app prompt, the author ranks Opus 5.5 first, Astra and Fable 5.1 tied, and GPT-6 Sol last for frontend design.

A screenshot claims OpenAI's official help center documentation now lists Astra Minor alongside astra, gpt-6 sol, and luna.

A user says a debugging task ran for over two hours on GPT-6 Sol with the 5-hour quota barely moving, and also shares a prompt for generating images of women.