Moonshot AI releases Kimi K3 on July 16 — a 2.8-trillion-parameter open-weight model competing with or beating GPT-5.6 and Claude variants in blind developer tests
Moonshot AI published Kimi K3 on 16 July with weights openly available: 2.8 trillion parameters, and competitive with or ahead of frontier closed models including GPT-5.6 and Claude variants in blind developer testing, reasoning and coding benchmarks. Rated as broadly on par with the best publicly available models of early 2026.
The substantive piece is the blind-test framing. Benchmark parity between open and closed weights has been claimed repeatedly and has usually dissolved under scrutiny of the evaluation setup. Blind developer preference is a harder result to dismiss, because it measures the thing buyers actually care about rather than the thing leaderboards measure.
The regulatory timing is not incidental. The AI Act's open-source carve-out excuses public-weight providers from two of four core obligations — but not from publishing a training-data summary. A 2.8T-parameter open-weight release lands directly into that requirement two weeks before enforcement powers activate.
The Neuron — Everything That Happened in AI Today → · Ink & Algorithms — AI Weekly Pulse — AI model releases July 2026 →