Kimi K3 Just Became the Largest Open AI Model Ever But “Free” and “Open” Both Need an Asterisk
Summary
Moonshot AI, a Beijing startup, released Kimi K3 on July 16, 2026, a 2.8-trillion-parameter Mixture-of-Experts AI model. This system is touted as the first "open 3T-class" and the largest open-weight model announced to date, quickly ascending benchmark leaderboards alongside models from OpenAI and Anthropic. However, the "free and open source" narrative surrounding Kimi K3 requires clarification. While Moonshot AI has promised full weight release by July 27, 2026, the model is currently only accessible via their proprietary API or website. Furthermore, it is not free, with API access priced at \$3 per million input tokens and \$15 per million output tokens.
Key takeaway
For Machine Learning Engineers evaluating new large language models, understand that Kimi K3's "open-weight" status is a future promise, not immediate availability for local deployment. You should factor in the API costs of \$3 per million input tokens and \$15 per million output tokens for any integration plans. Monitor Moonshot AI's official channels for the promised full weight release after July 27, 2026, before committing to a "free and open source" development strategy.
Key insights
The launch of Kimi K3, a 2.8T-parameter model, redefines "open-weight" AI while challenging "free" access expectations.
Principles
- "Open-weight" does not immediately mean downloadable.
- Large parameter counts correlate with benchmark performance.
- "Free" access often implies API costs.
In practice
- Access Kimi K3 via Moonshot AI's API.
- Evaluate API costs for large-scale integration.
- Monitor for full weight release post-July 27, 2026.
Topics
- Kimi K3
- Moonshot AI
- Large Language Models
- Mixture-of-Experts
- Open-weight Models
- AI Model Benchmarks
- API Pricing
Best for: CTO, VP of Engineering/Data, Investor, AI Scientist, Machine Learning Engineer, Director of AI/ML
Related on AIssential
See Counsel's argued verdicts on the open AI decisions leaders are weighing →
Editorial summary, takeaway, and curation by AIssential. Original article published by Artificial Intelligence on Medium.