A new concept I've been working on

I modded an open source kart racing game so the karts are driven by AI models instead of humans. Fable 5, Opus 5, GPT-5, and Gemini 3 Flash are each calling their APIs hundreds of times during this race to decide strategy, fire items at each

Jesse WaitesAug 14

Opus 5 rewrote the entire Bun runtime, 100,000 lines of Zig, into Rust

One prompt. 11 days. Thousands of parallel agents. And it's already in production inside Claude Code. Boris estimates a strong human team would've needed over a year 👇

RubenAug 141

I wanted to see how far Grok 4.6 would take a simple creative prompt.

Same instruction. Opus 5, GPT-5.6 Sol, and Qwen 3.8 Max joined in. Grok turned the banana into a tiny island. The different interpretations were way more interesting than I expected.

FHILY👑Aug 143

Grok 4.6 beats Claude Fable 5, Claude Opus 4.8, Gemini 3.1 Pro, GPT-5.5 and GPT-5.6 Sol on EEBench, an electrical-engineering benchmark.

DogeDesignerAug 14248