Apparently, pairing Opus 5 with Claude Design makes the results even more impressive.
GPT-5.6 Sol vs Opus 5. All in Three.js
Opus 5 scored 4X higher on the benchmark designed to be ungameable
ARC Prize ran the test nobody at a lab administers themselves: independent, third party verification on ARC AGI 3 - a benchmark built specifically so pretraining can’t help, only real-time reasoning in a
When you run the anonymous model "Ox Alpha" using OpenCode
When you run the anonymous model "Ox Alpha" using OpenCode, you can completely reproduce the basic behavior of MineCraft (voxel drawing/viewpoint movement/block placement) in one prompt. Now you can use OpenCode for free and without limit 🔥 Summary of surprising verification results ・Complete with only one HTML with one prompt ・Block physics that moves smoothly at 60fps…
You have reached the end of the archive
All of Claude Opus 5