Kimi K3: Moonshot AI Ships the Largest Open Model Yet
2 min readKimi K3, the newest model from Beijing-based Moonshot AI, launched on July 17 as the largest open-weight AI model the industry has seen. With 2.8 trillion parameters and native support for images and up to one million tokens of context, it lands as China’s boldest attempt yet to close the gap with the leading American labs.
What Moonshot Released
Moonshot AI is the company behind the Kimi chatbot, and Kimi K3 is its most ambitious release to date. At 2.8 trillion parameters, Moonshot says it is the world’s largest open-source model by scale. Unlike the flagship systems from OpenAI and Anthropic, Kimi K3 ships as an open model, meaning developers can download the weights and run or fine-tune it themselves rather than renting access through an API.
The timing was deliberate. Kimi K3 arrived on the opening day of the 2026 World Artificial Intelligence Conference in Shanghai, an event drawing more than 1,100 companies and over 300 product debuts. Chinese commentators framed the release as another “DeepSeek moment,” a reference to the low-cost model that rattled Western markets in early 2025.
How It Stacks Up
On raw benchmarks, Kimi K3 still trails the very top tier. According to reporting from CNBC and France24, it sits behind Anthropic’s Claude Fable 5 and OpenAI’s GPT-5.6 on overall performance. It did, however, beat Claude Opus 4.8 and GPT-5.5 on several tests, including coding and general agent tasks. For an openly downloadable model, matching systems that sit just behind the frontier is a notable result.
Why It Matters
Open models change the economics of AI. When a capable model is free to download, startups and researchers can build on it without paying per token, and they can run it on their own hardware for privacy or cost reasons. A 2.8-trillion-parameter open release also pressures closed labs to justify their pricing and pushes the open ecosystem closer to the commercial leaders.
The broader signal is geopolitical. China’s labs are shipping frontier-adjacent models on an open-weight strategy while the biggest US systems stay closed. Watch whether Kimi K3’s benchmark claims hold up under independent testing, and whether Western developers adopt it at scale.