Moonshot AI releases 2.8T-parameter Kimi K3, largest open-source model
Moonshot AI, a Beijing-based startup backed by Alibaba, released Kimi K3, a 2.8-trillion-parameter open-source model that benchmarks show performs competitively with top proprietary systems from Anthropic and OpenAI. The release, timed ahead of the 2026 World AI Conference in Shanghai, represents a significant escalation in the global AI race and marks a comeback for Moonshot after losing market position to DeepSeek over the past 18 months. Full model weights are scheduled for release on July 27, with the model already accessible via kimi.com.
TL;DR
- Kimi K3 contains 2.8 trillion parameters, roughly 75 percent larger than DeepSeek's V4 Pro at 1.6 trillion parameters
- The model features a 1-million-token context window, native visual understanding, and an always-on reasoning mode called thinking mode
- Benchmark results place K3 third on GDPval-AA v2 behind Claude Fable 5 Max and GPT-5.6 Sol Max, and second on AA-Briefcase agentic benchmark
- API pricing is $3 per million input tokens and $15 per million output tokens, with cached inputs at $0.30 per million, and the model is compatible with OpenAI SDK
Why It Matters
This release signals that open-source AI development is reaching parity with proprietary frontier models in both scale and performance. The 2.8-trillion-parameter scale and competitive benchmark results challenge the narrative that only well-funded U.S. labs can build cutting-edge systems. The timing and geopolitical context underscore intensifying competition between Chinese and Western AI companies.
Business Impact
Developers and enterprises now have access to a high-performance open-source alternative at mid-tier pricing, reducing lock-in to proprietary platforms from OpenAI and Anthropic. The OpenAI SDK compatibility lowers integration friction for existing deployments. The promotional pricing through August 12 creates a window for cost-sensitive organizations to evaluate the model at reduced rates.
Key Implications
- Open-source models are closing the performance gap with proprietary systems, potentially disrupting the pricing power of commercial AI labs
- Moonshot AI's recovery through a major release suggests the Chinese AI ecosystem remains competitive despite DeepSeek's earlier market dominance
- The 1-million-token context window and state-of-the-art BrowseComp score of 91.2 indicate advances in long-horizon reasoning and information retrieval tasks
What to Watch
Monitor adoption rates and developer feedback once full weights release on July 27, particularly whether the model's performance holds up in production workloads. Track whether the promotional pricing drives significant API usage and whether Moonshot can sustain competitive positioning against future releases from DeepSeek, OpenAI, and Anthropic. Watch for any regulatory or export restrictions that could affect distribution of the open-source weights.
Subscribe to the newsletter
The latest stories and analysis, delivered to your inbox.
Free. No spam. Unsubscribe any time.
