Moonshot AI unveiled Kimi K3 at the World Artificial Intelligence Conference (WAIC) in Shanghai on July 17, and it could reshape the open-source AI landscape. The model contains 2.8 trillion parameters and will be released as open-source on July 27, making it the world’s first freely downloadable model in the three-trillion-parameter class.
What Kimi K3 Can Do
Moonshot says Kimi K3 is its “most capable flagship model to date,” with strong performance across coding, knowledge work, and reasoning tasks. In internal benchmarks, Kimi K3 “demonstrated frontier-level performance,” consistently outperforming other tested models. However, Moonshot acknowledges that K3 “still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol.”
The model is natively multimodal and supports a 1-million-token context window, designed for long-horizon coding and knowledge work. Moonshot positions it for agentic coding tasks, multi-step reasoning, and document analysis across extended conversations.
The Open-Source Angle
When Kimi K3 launches on July 27, it will be the first open-source model at the three-trillion-parameter scale. Developers will be able to download, run, and customize it freely. This is significant because most frontier-class models remain locked behind API access from companies like OpenAI, Anthropic, and Google.
The move follows a pattern set by Meta’s Llama series and Mistral’s models: release a capable open model to build a developer ecosystem, then monetize through cloud services and larger proprietary models. For Moonshot, Kimi K3 serves as both a technical showcase and a business strategy.
The Founder: Yang Zhilin
Moonshot’s co-founder and CEO Yang Zhilin earned his PhD at Carnegie Mellon University in 2019, where he made contributions to the machine-learning field. His former advisor, Russ Salakhutdinov (also a former director of AI research at Apple), praised the K3 release as “a huge win for the open-source community.”
Zhilin is known for his love of Pink Floyd and built Moonshot from a small Beijing lab into one of China’s most valued AI startups. The company’s previous models, including Kimi K2 and K2.5, had already gained traction among developers for their strong reasoning capabilities.
Geopolitical Context
The timing of Kimi K3 matters. At WAIC, President Xi Jinping outlined China’s vision for global AI leadership, emphasizing open-source development as a key strategy. Moonshot’s decision to release K3 openly aligns with China’s push to build AI influence through open ecosystems rather than proprietary lock-in.
For US-based AI labs, this creates pressure. An open model at the 2.8T parameter scale that matches or approaches proprietary model performance means developers no longer need to rely on American APIs. Mira Murati’s Thinking Machines Lab released its own open-weight model (Inkling, 975B parameters) on the same day, but Kimi K3 dwarfs it in scale.
The competition extends beyond model size. Moonshot’s Kimi platform offers direct API access at competitive pricing, and the K3 model integrates with tools for agentic coding, autonomous task execution, and parallel processing through its “Swarm” and “Goal” features.
When Can You Try It?
Kimi K3 is available now through the Kimi API at api.moonshot.ai. The open-source weights will be released on July 27 through Moonshot’s GitHub and Hugging Face repositories. The model can also be tested through the Kimi chat interface at kimi.com.
FAQ
What is Moonshot AI Kimi K3?
Kimi K3 is a 2.8 trillion parameter AI model from Chinese startup Moonshot AI, unveiled at the World AI Conference in Shanghai on July 17, 2026. It will be released as open-source on July 27.
How does Kimi K3 compare to ChatGPT and Claude?
Moonshot acknowledges that Kimi K3 “still trails the most powerful proprietary models, Claude Fable 5 and GPT 5.6 Sol.” However, it demonstrates frontier-level performance and outperforms other tested models across coding, reasoning, and knowledge tasks.
When will Kimi K3 be available as open-source?
The open-source weights for Kimi K3 will be released on July 27, 2026. It will be downloadable from Moonshot’s GitHub and Hugging Face repositories.
What is the context window for Kimi K3?
Kimi K3 supports a 1-million-token context window, designed for long-horizon tasks like extended coding sessions and multi-document analysis.
Who founded Moonshot AI?
Moonshot AI was co-founded by Yang Zhilin, who earned his PhD from Carnegie Mellon University in 2019. His former advisor described the K3 release as “a huge win for the open-source community.”
