What is Kimi K3?
Kimi K3 is Moonshot AI's open-weights flagship, released in July 2026 as the successor to the K2 line. It belongs to the Kimi family — native multimodal models tuned for agentic coding, long-document reasoning, and autonomous multi-step work. Overchat AI gives you direct access without needing a Moonshot account or a Chinese phone number.
Moonshot AI, the team behind Kimi, was founded in Beijing in March 2023 by Yang Zhilin, Zhou Xinyu, and Wu Yuxin — all Tsinghua alumni. Yang earned his PhD at Carnegie Mellon and worked on NLP at Meta and Google Brain before co-founding the lab, which has since become one of China's most visible open-source AI labs.
K3 is Moonshot's strongest release to date, trading places with closed frontier models like Claude Fable 5 and GPT-5.6 Sol on coding, agent, and knowledge benchmarks — while shipping its weights publicly.
Kimi K3 Features
A 1M-token context window lets K3 read and reason over long codebases, legal filings, or research archives in a single pass, and it can write back up to the full million tokens when a task calls for it. Kimi Delta Attention — a hybrid linear attention design — keeps inference costs in check even when the conversation stretches far beyond what typical chatbots hold.
Against the K2 line, K3 scales from a trillion parameters to 2.8 trillion, quadruples the context window to a full million tokens, and adds native visual understanding alongside a configurable reasoning-effort setting.
Kimi K3 ships with open weights under the Kimi K3 License, so you can pull them from Hugging Face and run the model on your own hardware. If you'd rather skip the setup — and 2.8 trillion parameters is a lot of hardware — Overchat AI lets you chat with it instantly, with no Moonshot API key or China-region account required.
Through Moonshot's own API, K3 costs $3 per million input tokens and $15 per million output tokens, and exposes tool calling with dynamic tool loading for agent workflows. On Overchat AI you don't pay per token — K3 comes with your plan alongside the other top models.
What makes Moonshot stand out is how much it puts in the open. The weights and the benchmark tables are public, so researchers and developers can inspect, fine-tune, and extend K3 instead of treating it as a black box. That openness matters for critical work where you want a coding partner you can audit end to end.
.svg){kind=icon}
