GPT-6 Sol leaks, screenshots and release rumors

Higgsfield made two versions of a samurai showdown to compare Claude Opus 5.5 and GPT-6 Sol in animated filmmaking.

Higgsfield shows a notebook sketch turned into an animated sequence of 2D characters in a 3D environment, comparing Opus 5.5 and GPT-6 Sol.

A joking reply to @OpenAI says that even a homemade benchmark won't help GPT-6 Sol beat Opus 5.5.

Higgsfield posted a video comparing GPT-6 Sol and Claude Opus 5.5 on the same ink-wash animation task.

Given the same sunset ocean prompt, GPT-6 Sol finished in 44 seconds for $0.07, far faster than Claude and Astra, and the author asks which result was best.

In the Base ecosystem weekly roundup, No. 1 Venice handles 250B tokens a day and added anonymous access to GPT-6 Sol and Luna.

A paying Codex user compares Opus 5.5, Astra and GPT-6 Sol, saying Sol is only cheaper rather than a real upgrade and that usage runs out too quickly.

Higgsfield shares a second comparison video in which GPT-6 Sol and Claude Opus 5.5 animate notebook doodles.

Four new models set eleven new Pareto points this week: GPT-6 Sol (max) scores 48 at $1.06 per task, while Opus 5.5 leads at 58.

On the same prompt, Opus 5.5 cost $1.27 and took 9m 35s, GPT-6 Astra $0.63 and 5m 4s, GPT-6 Sol $0.08 and 50s, and Luna under a cent and 43s.

After a muted reception for GPT-6 Sol and Luna, the post predicts next week's DevDay could unveil a mysterious "Aeon" bot and a new kind of agentic experience.

A MAX 20x user says a full day across two projects left 91% of quota, using Sol as lead, Luna for simple tasks and Astra only for the hardest problems.