Kimi K3

Kimi K3 — explained in more detail

Moonshot AI released Kimi K3 on July 16, 2026 as its new flagship; the open weights went public on July 26, 2026 — a day ahead of the announced target. It is a Mixture-of-Experts (MoE) model with 2.8T total parameters, of which about 104B are activated per token. The model has native vision and a context window of 1M tokens.

Example / Practical use

At 2.8T parameters, Kimi K3 is the first open-source model to reach the 3-trillion-parameter class, and the largest open-weight model released to date — larger than DeepSeeks V4 Pro (1.6T) and Zhipu AIs GLM 5 range (744B). Moonshot made the weights available for unrestricted public download; developers can freely deploy, adapt and self-host the model. In individual benchmarks, K3 beat Western frontier models such as Claude Fable 5, for example in frontend code comparisons. For teams with their own infrastructure, the free access is the central practical advantage.

Distinction

Kimi is the model range of the Chinese company Moonshot AI. Unlike proprietary US models from OpenAI, Anthropic or Google, Kimi K3 stands for frontier-level open weights. Compared with other open models such as DeepSeek V4 or GLM 5, K3 sets itself apart through sheer size and native vision capability.

More on this topic

See everything in one place:Kimi→

Encyclopedia · 12 Sept 2026Kimi (Moonshot AI) — the open model line at a glanceKimi by Moonshot AI as an open line: positioning, use profile and how to choose. Flagship Kimi K3 with a very large context window.GlossaryKimi K2.5Kimi K2.5 is an open model from Moonshot AI released in January 2026. It is a native multimodal Mixture-of-Experts model with around 1 trillion parameters (32 billion active), a 256,000-token context and agentic capabilities.GlossaryKimi K2 ThinkingKimi K2 Thinking (November 2025) is Moonshot AI's open reasoning-agent model — a trillion-parameter MoE with a 256K context that runs long thinking and tool-use chains on its own.