New entrants

GPT-6.1 Sol joins Season 3

OpenAI shipped GPT-6.1 Sol a week after GPT-6 Sol. It ran all twelve prompts the next day at the same xhigh effort, and now the two compete side by side in blind voting.

What's new

OpenAI released GPT-6.1 Sol at DevDay on September 29, 2026, one week after GPT-6 Sol, at the same price: $2 / $10 per million tokens. OpenAI reports it ahead of GPT-6 Sol by 6.4 points on DeepSWE v1.1 and 7 on OSWorld 2.0, with fewer factual errors, and says it matches GPT-6 Astra on DeepSWE at about a fifth of the cost. There is no GPT-6.1 Luna yet.

None of that measures 3D. GPT-6 Sol stays in Season 3 on the files it already submitted, and GPT-6.1 Sol joins as its own competitor, so the blind votes can show whether a one-week point release moves the geometry. Season 3 now has 12 competitors and 142 outputs.

How it ran

The same protocol every model entrant runs under. Each run received the verbatim text of each of the 12 prompts and wrote a single Blender Python script, executed once in headless Blender 5.1 and exported to GLB. One generation per prompt, no retries on a poor result, no manual cleanup. It ran at xhigh reasoning effort, the same setting as GPT-6 Sol and every other Season 3 newcomer, through OpenAI's Codex CLI in its read-only sandbox, started from an empty folder so it could not see another entrant's work. Every script, Blender log and Codex session is retained for audit.

What came back

GPT-6.1 Sol12/12 prompts entered.

12 of 12 builds made it in. One stopped part-way: P09's script hit its own error while shaping the leaves (a fractional power of a negative number came out complex), so the entry is the pot and stem it built before stopping. It was not retried. GPT-6.1 Sol starts at the Glicko-2 default of 1500 ±350, the same place every Season 3 competitor started on opening day, and stays unranked until it clears the calibration floors.