Moonshot AI’s official Kimi API documentation currently lists Kimi K2-series models, not a Kimi K3 release. Third-party posts describe claims of a 2.5 trillion-parameter, 1 million-context Kimi K3 as leaks, with no official model card, weights, pricing, or benchmarks yet available.
Moonshot AI currently lists Kimi K2-series models in its public Kimi API documentation, but the company has not published an official Kimi K3 model entry or release announcement.
Moonshot AI’s Kimi API Platform model list names several available models, including kimi-k2.7-code, kimi-k2.7-code-highspeed, kimi-k2.6, and kimi-k2.5. According to the Kimi API Platform documentation, these listed models show a 256k context length. The same public catalog does not list a Kimi K3 model ID, pricing, weights, or model card.
Moonshot AI’s official homepage also continues to present Kimi K2.6 as its current flagship in the visible materials referenced here. Moonshot AI’s public research listings cited in the source material stop at Kimi K2.6, dated April 20, 2026, and do not show a visible Kimi K3 announcement.
That means the central public evidence available from Moonshot itself points to K2-series availability, not a confirmed K3 launch.
The most widely circulated claims around Kimi K3 describe a model with 2.5 trillion parameters, a 1 million-token context window, and open-source availability. Those claims are not confirmed by Moonshot AI in the official sources provided.
AI Reiter, in a post titled “Kimi K3 Leak: Launch Signals, Specs, and What to Use,” characterizes Kimi K3 as not officially released. AI Reiter treats the reported July 15 launch page, 2.5 trillion-parameter claim, and 1 million-context claim as leaks rather than confirmed specifications.
Kie.ai similarly describes Kimi K3 as unreleased. In its article “What Is Kimi K3? Moonshot’s 2.5T, 1M-Context Flagship,” Kie.ai says there is no official model card, benchmark set, pricing, or weights available, and labels the 2.5 trillion-parameter and 1 million-context figures as unconfirmed community leaks.
Those third-party accounts are useful for tracking what is being discussed, but they do not substitute for an official release note, model card, repository, pricing page, or API documentation entry from Moonshot AI.
Large language model announcements are often evaluated on several concrete artifacts: an official model name, documented context length, benchmark methodology, deployment options, license terms, pricing, and, for open models, downloadable weights or a repository. In the sources reviewed here, those artifacts are available for Moonshot’s listed Kimi K2-series models, but not for Kimi K3.
The difference is especially important because the rumored figures would be significant if verified. A 2.5 trillion-parameter model and 1 million-token context window would be major technical claims. Open-source availability would also require specific licensing and weight-release details. None of those details appear in Moonshot AI’s public model list or homepage materials cited here.
For now, the most accurate reading is that Kimi K3 is a rumored or anticipated model, not a confirmed public release. Moonshot AI’s official Kimi API Platform remains the primary source to watch for any model ID, context length, pricing, or access details. Until Moonshot publishes those details, claims about Kimi K3’s size, context window, and release status should be treated as unverified.
Moonshot AI currently lists Kimi K2 series models in its public Kimi API documentation, but the company has not published an official Kimi K3 model entry or release announcement.
What is confirmed Moonshot AI’s Kimi API Platform model list names several available models, including kimi k2.7 code , kimi k2.7 code highspeed , kimi k2.6 , and kimi k2.5 .
According to the Kimi API Platform documentation, these listed models show a 256k context length.
Continue reading
