Test command of 3 agents Gemini 3.8 Flash / Claude Fable 5.1 / GPT-6 Astra with the same commands created with Claude OPUS 5 as follows.
As a result, the GPT-6 Astra wins in terms of gaming aesthetics. (According to the clip, you'll only get this because you didn't let it think for itself.) The second best work is Claude Fable.
GPT 6 Astra one shotted a bug in BridgeMind that no model could solve in 6 months.
Fable 5 failed. Fable 5.1 failed. Opus 5, Sol, Grok 4.6, all failed. Astra fixed it on the first prompt. Then it rebuilt my office in Blender from photos. Built a racing game in 13 minutes that
Fable 5.1 is here, and here's why it's better than what came before it
It solves more coding problems than Fable 5 or Opus 5, and hits state of the art on trading intuition An investment firm, Millennium, used it to find the cause of a rare crash in their systems that none of
Someone gave GPT-6 and Fable 5.1 access to Canva.
One smoked the other (it's not even close). Claude Fable 5.1 is on the left. GPT-6 Astra is on the right. The gap is not small. It comes down to one test (ARC-AGI 3). It drops a model into a game with no instructions and
You have reached the end of the archive
All of Claude Opus 5