GPT-6 Sol leaks, screenshots and release rumors

eidzoku says GPT-6 Sol feels more like 5.2/5.3 Codex than 5.6, struggles to read intent and casual chat, and is considering Claude Max to use Sol only as a subagent.

Arena launched a redesigned leaderboard overview that gathers live scores and first capability impressions for new models like GPT-6 Sol and Claude Opus 5.5 across categories.

The author gave Opus 5.5 and GPT-6 Sol the same photo and asked each to draw them in Paint using computer use, with results funny enough to lose it over, plus a video.

A short post predicts that using Opus 5.5 as the main controller with GPT-6 Sol running subagents will soon become the mainstream setup.

A question post asks whether GPT-6 Sol and Luna will ever be made available in the regular version of ChatGPT, and when.

A tester says GPT-6 Sol is more natural and humorous than Astra but has strong guardrails and consistently short replies similar to Opus 5.5, and laments that model personalities are being flattened.

The poster says GPT-5.6 Sol feels warm and human while GPT-6 Sol is cold and detached in work settings, and worries that merging chat with work would cost ChatGPT its conversational magic.

A video says OpenAI released GPT-6 Sol at half the price roughly 90 minutes after Anthropic launched Claude Opus 5.5, with early independent tests split on which is better.

A user compared two new models in the same week: GPT-6 Sol burned 40% of their limit in two hours, while Opus 5.5 used 12% in three, despite Sol being marketed as efficient.

A third-party benchmark screenshot shows GPT-6 Sol slightly better than 5.6 Sol, Grok 4.7 slightly worse than 4.6, and GPT-6 Luna performing poorly.

The poster says they keep going back to GPT-5.6 Sol xhigh after using GPT-6: it is cheaper but needs more corrections and babysitting, while 5.6 just gets things done.

The developer says they use Opus 5.5 as the main agent and sometimes stack it with GPT-6 Sol, and use Opus 5.5 plus GPT Astra for image tasks.