On July 16, 2026, Moonshot AI introduced Kimi K3, its most capable model to date: a 2.8 trillion parameter system with native vision capabilities and a 1-million-token context window, which the company calls the world’s first open 3T-class model. Moonshot says K3 demonstrates frontier-level performance across its evaluation suite, consistently outperforming other tested models, while acknowledging that overall performance still trails the strongest proprietary systems, Claude Fable 5 and GPT-5.6 Sol.
Architecturally, K3 is built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes), two updates aimed at improving information flow across sequence length and model depth, combined with a scaled-up Mixture of Experts design that activates 16 of 896 experts under a Stable LatentMoE framework. Moonshot claims these changes deliver roughly 2.5x better scaling efficiency than Kimi K2, converting compute into capability more effectively. The company notes that for nine of the past twelve months, Kimi models have set the upper bound of open-model sizes.
Kimi K3 launched immediately on Kimi.com, Kimi Work, Kimi Code, and the Kimi API, running at max thinking effort by default with lower-effort modes to follow. Crucially for the open ecosystem, Moonshot committed to releasing the full model weights by July 27, 2026, alongside a technical report, and said it was working with inference partners and open-source maintainers to prepare the rollout. The release intensified the price and capability pressure on US labs during a two-week stretch that also saw GPT-5.6 and Grok 4.5 ship.