Moonshot AI, a Beijing-based startup backed by Alibaba, has released Kimi K3, a 2.8-trillion-parameter model that it claims is the world’s largest open-source AI model. The launch, timed ahead of the 2026 World Artificial Intelligence Conference in Shanghai, marks a major moment in the global AI race and a comeback for Moonshot after losing ground to DeepSeek over the past 18 months. Full model weights are expected on July 27, while the model is already accessible through kimi.com.
Kimi K3 features a 1-million-token context window, native visual understanding, and an always-on “thinking mode.” It is built on Moonshot’s own Kimi Delta Attention and Attention Residuals architectures. The model is also compatible with the OpenAI SDK, and its API pricing is set at $3 per million input tokens and $15 per million output tokens.
Benchmark results suggest Kimi K3 performs at or near the level of leading proprietary systems from Anthropic and OpenAI. It ranked highly on tests such as GDPval-AA v2, AA-Briefcase, BrowseComp, and frontend coding leaderboards, often finishing just behind or ahead of top competitors like Claude Fable 5 Max and GPT-5.6 Sol Max.
Moonshot also showcased ambitious demos, including a 48-hour autonomous chip-design task and rapid astrophysics research reproduction, highlighting its focus on long-horizon agentic capabilities. The release is also a geopolitical signal: by open-sourcing such a massive model, Moonshot aims to strengthen China’s role in the global open-source AI ecosystem.
Article timeline
Build this article’s timeline
Extract dated events, key actors, and open questions using the same timeline analysis available in Signal.
Moonshot AI, a Beijing-based startup backed by Alibaba, has released Kimi K3, a 2.8-trillion-parameter model that it claims is the world’s largest open-source AI model.
🤖 AI prefill
📅WhenWhen did this happen, or what is the timeframe?
📍WhereWhere did this occur?
💡Why
Centers on Kimi K3, 2.8-trillion-parameter model, and open-source AI model
🤖 AI prefill
Login to answer and earn XP for each gap you fill.
Comments
Login to view and post comments.