Moonshot AI has released Kimi K3, a new artificial intelligence model with a substantial 2.8 trillion parameters. This advanced model is engineered with a one-million-token context window, a feature designed to significantly enhance its capabilities in complex reasoning, long-horizon coding projects, and knowledge-intensive work.
The full release of the model’s weights is anticipated on July 27. However, users can already access Kimi K3 through existing Kimi services and API routes. The model incorporates native visual understanding capabilities and features internal architectural innovations, including what the company describes as Kimi Delta Attention and Attention Residuals.
Independent benchmark assessments have positioned Kimi K3 favorably, placing it in proximity to leading frontier models, particularly on tasks involving extended context windows and interface development. These evaluations also noted that operating such a large model locally necessitates considerable computing resources.
In terms of cost, Kimi K3 is priced at $3 per million input tokens and $15 per million output tokens. Cached input is available at a rate of $0.30 per million tokens. The timing of this launch appears strategic, preceding the 2026 World Artificial Intelligence Conference in Shanghai and contributing to a growing trend of substantial open-model releases originating from China.
The development and release of Kimi K3 by Moonshot AI represent a significant advancement in the field of artificial intelligence, pushing the boundaries of what is possible with large language models. The model’s extensive context window and sophisticated architecture suggest potential applications across a wide array of demanding computational tasks. The company’s strategy to provide early access while preparing for a full weight release indicates a phased approach to market integration, allowing developers and researchers to begin exploring its capabilities.
While the benchmark results are promising, the requirement for significant local computing power for large-scale operations suggests that widespread adoption may depend on the availability of robust infrastructure. The pricing structure, with a notable difference between input and output token costs, will also be a key factor for businesses and researchers considering its implementation. The competitive landscape, marked by a wave of Chinese open-model releases, suggests an accelerating pace of innovation in the global AI sector.
This release from Moonshot AI is part of a broader trend in the AI industry towards increasingly powerful and versatile models. The integration of visual understanding and novel attention mechanisms could unlock new possibilities for AI-driven analysis and creation. As the July 27th full weight release approaches, the AI community will be closely watching how Kimi K3 is integrated into various applications and research endeavors, potentially setting new benchmarks for performance and capability in the months and years to come.
