Chinese AI company Moonshot AI has released Kimi K3, which it calls its most capable model yet. The company announced the launch on July 16, 2026, according to MarkTechPost.
The headline number is size: Kimi K3 has 2.8 trillion parameters. Moonshot says it rivals two of the strongest closed models on the market, Opus 4.8 and GPT 5.5, according to Techmeme. The company also plans to release the model's weights — the underlying trained values that let others run and build on it — by July 27, per Techmeme.
Under the hood, MarkTechPost reports that Kimi K3 is an open "mixture-of-experts" (MoE) model, a design that routes each request to a small slice of the network rather than firing up the whole thing. In K3's case, it activates 16 of 896 experts at a time. The model is built on techniques Moonshot calls Kimi Delta Attention and Attention Residuals, and supports a 1-million-token context window, meaning it can work over very long documents in a single pass.
On independent benchmarks, Kimi K3 launched at a score of 57 on the Artificial Analysis Intelligence Index, according to a report circulated via LinkedIn and Google News. The launch also drew heavy attention on Hacker News, where a post titled "Kimi K3: Open Frontier Intelligence" gathered more than 1,000 points and 630 comments.
Why it matters: An openly released model claiming parity with the best closed systems from leading labs could put frontier-level AI directly into the hands of researchers and developers worldwide, intensifying competition over who controls advanced AI.