GPT-6 Sol leaks, screenshots and release rumors

In an internal OpenDesign test, Opus 5.5 and GPT-6 Sol were given the same prompt to build a scrollable 3D city website; this post shows the Opus 5.5 result.

The post compares Sol at $2/$10 with Opus 5.5 at $4/$20, noting Opus claims an edge in agentic coding while Sol pushes harder on cost, and asks which readers would pick.

A meme jokes that Anthropic shipped Opus 5.5, OpenAI shipped GPT-6 Sol and Luna, and SpaceXAI shipped Grok 4.7, while Google stayed quiet.

A user says Astra High worked for about 20 minutes on their Plus plan while GPT-6 Sol worked for 50, arguing that is not unlimited.

The author ran Claude Opus 5.5 and GPT-6 Sol high on the same day and found the $20 Claude Pro tier close to the $100 Codex tier in productivity, with 65% and 57% of quota left respectively.

The post calls Sol a delusional disaster, says Luna is only decent for simple tasks, and argues Opus 5.5 is the new SOTA while criticizing OpenAI's marketing of smaller models.

A Japanese user asked GPT-6 Sol to delete a piece of meat that kept floating in the lower-left of a video, and it fixed the clip in one attempt; before-and-after versions were shared.

A short post says this is the golden age of software engineering, calling both GPT-6 Sol and Opus 5.5 amazing.

An Arabic discussion asks, after real testing, which of GPT-6 Sol and Opus 5.5 is stronger, which disappointed, which people rely on for work, and where Luna fits.

The author says switching Codex to GPT-6 Sol cut development time by 90% and token use by half with no Skills installed, and shares a minesweeper prompt plus a demo of the finished game.

Given the same prompt, Opus 5.5, Grok 4.7, GPT-6 Luna and Sol each built a little monster that follows the cursor and reacts to feeding; Luna finished first but its expressions fell apart.

A repost says Sol and Luna are live and lists token prices about 92% below official rates.