GPT-6 Sol leaks, screenshots and release rumors

A tester says GPT-6 Sol completes tasks with half the tokens and in a fifth of the time versus GPT-5.6 Sol, and with the 50% cost cut it works as a daily driver.

A side-by-side test of Sol and Luna on the same prompt with no harness finds both clearly below Opus 5.5, though the author notes the large price gap.

A short post noting that cancelling Claude was followed by the Opus 5.5 launch, and cancelling GPT was followed by GPT-6 Sol.

The video says Sol targets complex work and coding while Luna handles everyday bulk tasks like summarization, both sitting below Astra, with Opus 5.5 launching the same day and Grok 4.7 the day before.

A GPT Plus user says a first try of GPT-6 Sol in Work mode gave short, boring answers much like Astra, and that the model is not yet available in Chat mode.

A long comparison of Astra and Fable on coding finds Fable the favorite for most complex work, with a GPT-6 Sol follow-up promised once it is available.

The open-source Multi-Provider Orchestration project assigns tasks by model: Sol for implementation, Luna for routine work and Opus 5.5 for reviews.

A Japanese AI news brief says GPT-6 Sol and Luna have been released with API prices cut in half, and also mentions Opus 5.5 and a new Snapdragon chip.

Linear announced that GPT-6 Sol, GPT-6 Luna and Claude Opus 5.5 are now available in its coding sessions.

Across five drafts of 10 YouTube scripts scored blind by three AI judges, Sol ranked 6th of 94 models at $0.11 per script, with its effort setting spanning 124 Elo from default to max.

A user says Sol and Luna deliver GPT-5.6-level capability at half the cost, and suggests Astra for planning, Sol for implementation, and Luna for simple tasks.

iannuttall says a new benchmark feels closer to everyday model use and names GPT-6 Sol as the result he wants to see, an expectation rather than a test.