GitHub just dropped Project HydraFusion inside Copilot CLI.

One picker. Multiple models. Runtime picks the workflow — Single, Cascade, or Critique. Vs Claude Opus 5 (GitHub offline evals): • TerminalBench 2.1: +4.9 pts quality, 67% cheaper • DeepSWE: near-parity, 36% cheaper

The Raw Wire 🚨Sep 52

I asked Claude Opus 5 to paint the Captain America with JavaScript.

RooSep 52

I finally made a fully procedural space level designer that has taste, looks and usability.

Science + math by Fable 5.1 Execution by Opus 5. Powered by threejs. 🔽🔽🔽 ⚠️Disclaimer: What you see is pure math, no 3D models used and no textures, effects, or retouching applied.

Startracker 🔺Sep 516

Openai made GPT-6 astra public today and the benchmarks are making every other lab uncomfortable

That's not a tagline. that's the president of OpenAI telling reporters in person that he believes they've crossed the threshold here's what the benchmarks actually say ARC-AGI-3:

PolyBenderSep 5