GPT-6 Sol leaks, screenshots and release rumors

A question post soliciting impressions pairs Sol and Luna together, reflecting that both have entered discussion.

The post says Sol and Luna halve prices and move toward Astra's alignment and language style, with Sol beating Opus 5 on AutomationBench at about 9% of the cost but trailing GPT-5.6 Sol on some OSWorld 2.0 metrics.

Claire Vo posted a live comparison review of Opus 5.5 and GPT-6 Sol on YouTube, using Codex and ElevenLabs to fix the audio.

A Korean analysis says OpenAI's half-price Sol/Luna push value while Anthropic dropped Haiku for Opus 5.5, so the better subscription depends on what you use it for.

A user complains that OpenAI followed its best model, Astra, with GPT-6 Sol, which can't even outperform the previous generation.

The user says GPT-6 Sol feels like the previous generation pushed to its limit and made cheaper, while Anthropic's Opus 5 to 5.5 jump feels like a full generation ahead.

The author gave GPT-6 Sol and Opus 5.5 the same prompt to each build a game, says the gap was huge and Opus 5.5 won outright, with links to both games in the video.

A video shares Night Train benchmark results with Opus 5.5 the clear winner, and the author wants to test whether the newly available GPT-6 Sol and Luna can beat it.

An Artificial Analysis chart shows GPT-6 Sol and Luna halving the cost of their predecessors at similar Intelligence Index scores, driven by a roughly 50% cut in token pricing.

One poster says GPT-6 Sol has been pretty disappointing in practice, writing that they expected more, among the negative feedback after launch.

A Japanese daily roundup says GPT-6 Sol and Luna are out at roughly half the GPT-5.6 API prices, with better coding and alignment but only small benchmark gains.

OpenAI cut GPT-6 prices in half, with Sol at $2/$10 per million tokens beating Opus 5 on AutomationBench at 9% of the cost per task, plus improved writing style.