Kimi K3 Beat Claude at Coding. It Also Hallucinates 51% of the Time.

Thanks as always for the support — it's what makes the fact-checking possible. 🙏

Two things happened this month that don't sound like they belong in the same story. A Chinese lab released the largest open-weight AI model in history — Kimi K3, 2.8 trillion parameters — and in blind developer testing it beat Claude Fable 5 and GPT-5.6 Sol at frontend coding. Then the independent evaluators published a second number: the same model fabricates answers 51% of the time. Worse than its own predecessor.

Champion coder. Confident liar. Same model, same week.

Hype threads gave you the first half. Skeptics gave you the second. I fact-checked every major claim against primary sources and wrote the whole thing — what Kimi K3 actually is, which claims hold up, why "open weights" doesn't mean what most people think, and what a builder should actually do about it. 👇

📖 Read it: https://medium.com/@reactjsbd/the-biggest-open-ai-model-ever-just-beat-claude-at-coding-it-also-hallucinates-half-the-time-83b50de39023?sharedUserId=reactjsbd

If it's useful, a coffee keeps the research going. ☕❤️