Kimi K3: The New Frontier in Open-Weight AI

On July 16, 2026, Moonshot AI, a Beijing-based startup, released Kimi K3, a groundbreaking open-weight model with 2.8 trillion parameters. It is the largest publicly available model of its kind, representing a significant leap in AI research. The full weights are scheduled to be released by July 27, 2026, but the model is already accessible via API and app.

Pricing and Token Costs

Kimi K3 is priced competitively, positioning it between the cheapest models from Chinese labs and the more expensive top-tier models. The token costs are as follows:

Input tokens: $3.00 per 1 million tokens. For cached inputs (prompt caching), a 90% discount applies, reducing the cost to just $0.30 per 1 million tokens. Output tokens: $15.00 per 1 million tokens. The model supports a context of up to 1 million tokens, with pricing remaining flat across that range.

This pricing structure makes Kimi K3 an attractive option for extensive coding, knowledge work, and reasoning tasks, especially when large context windows are required.

Benchmark Performance

In benchmarks, Kimi K3 demonstrates impressive performance, bringing it close to the leading proprietary models. On the Artificial Analysis Intelligence Index, it scored approximately 57.1, placing it in the same tier as Claude Opus 4.8 and GPT-5.5. However, it trails behind the absolute leaders, such as Anthropic's Fable 5 and OpenAI's GPT-5.6 Sol.

Notably, Kimi K3 excels in coding and agentic tasks. It reached the #1 spot on the LMArena Frontend Code Arena, outperforming Claude Fable 5 and GPT-5.6 Sol in several front-end development categories. This underscores its strengths in practical software engineering.

Compared to other open-weight models, Kimi K3 is the leader, significantly surpassing DeepSeek V4-Pro (1.6 trillion parameters). It uses a Mixture-of-Experts (MoE) architecture with 896 experts (16 active per token) and introduces novel architectural components like Kimi Delta Attention (KDA) and Attention Residuals to improve efficiency and information flow over long sequences.

Open-Weight Leadership

Kimi K3 is considered the largest open-weight model ever released. While it technically only releases the weights, not the training data or pipeline, it is still a significant milestone. It is the strongest challenger to Western closed-source models in the open-weight ecosystem, offering unprecedented scale and performance to the AI community.

The release of Kimi K3 demonstrates that open-weight models can now compete with closed-source models, particularly in niches like front-end development. This could reshape the AI landscape by enabling more companies and researchers to access cutting-edge AI capabilities without relying on expensive proprietary services.

Conclusion

Kimi K3 is a groundbreaking open-weight model that sets new standards in size, performance, and accessibility. With competitive pricing, impressive benchmark results, and open-weight availability, it presents a serious alternative to the closed-source models from OpenAI, Anthropic, and Google. It is a major step toward democratized AI and is poised to influence the development of AI research and applications significantly.