GPT-6 Sol leaks, screenshots and release rumors

The author says tasks that take Claude Opus 5.5 over an hour are done by GPT-6 Astra in 15 minutes or less, so they often use Astra and Sol for time-sensitive work.

A meme-style post jokes that after Anthropic and OpenAI each released new models, Google said it is taking AI to space.

A Portuguese post says OpenAI launched GPT-6 Sol and Luna with permanent pricing of $0.10/$0.50 per million tokens, that Luna matches GPT-5.6 Sol at 1/100 the cost, and that Opus 5.5 cut Anthropic's prices the same day.

On the $200 plan, a full day of heavy Sol use took about 8% of the weekly limit, and the author says the two models work better together than as a contest.

A user argues that after the launches of GPT-6 Sol and Opus 5.5, even Elon Musk realized Grok 4.7 was a failure.

The user says they liked Astra but finds GPT-6 Sol a joke, and has moved to Opus 5.5, which they say is performing strongly.

After several hours of testing, the user says Sol falls well short of Opus 5.5 on frontend and backend work, with generous usage as its only bright spot, and plans to return to Opus.

The ICP platform now lets users pick a model for building directly on the Internet Computer, with options including Opus 5.5, Fable 5.1, GPT-6 Astra, GPT-6 Sol and Kimi K3.

The author gives an equivalence table for Opus 5.5 low, medium and high effort against GPT-6 Sol and Astra reasoning levels, concluding that switching effort is more common than switching models.

A comparison video of two models building a 3D game from the same prompt, with viewers asked to vote on which one did better.

The same prompt was run through Opus 5.5, GPT-6 Astra, Sol and Luna, and the author says running Sol sub-agents under an Astra main agent is good value.

A one-line take arguing that GPT-6 Sol feels like the benchmark model that Opus 5 once was.