Two models same cezanne brief one just filled a template
A reviewer gave Opus 5 and GPT-5.6 the identical brief - interactive 3D journey through Mont Sainte Victoire, five versions, same painter The gap wasn't - It was in what each model decided actually mattered Opus 5
Same model. Same benchmark.
Same day. The score jumped from 62.7% to 99.9% — and nothing about the model changed. GPT-6 Astra just solved ARC-AGI-3, and everyone's crediting the model. The real story is the harness around it. 5:50 – the baseline: GPT-5.5 <1%, GPT-5.6 Sol 7.8%,
Comparison of Gemini 3.8 Flash with ChatGPT SOL, Opus 5 and Kimi K3
In this test, Gemini 3.8 Flash does not perform very well in the production of 3D content. This model provided such a weak output in about 2 minutes; Interestingly, in Goal Mode, the execution of the same task took about 30 minutes
Claude Opus 5 is INSANE
Look what I’ve just built.
You have reached the end of the archive
All of Claude Opus 5