Kimi K2 Turbo Edition AI Model Accelerates Further, Reaching Output Speed of 100 Tokens per Second
-
Recently, Moonshot AI released the latest updates on the Kimi K2 Turbo Edition AI model. Thanks to the engineers' efforts, the output speed of the Kimi K2 turbo-preview model has significantly improved, now achieving a stable output speed of 60 tokens per second, with peaks reaching 100 tokens per second. This advancement marks a substantial increase in the model's data processing efficiency, allowing users to experience smoother services.
Kimi K2 is a foundational model based on the MoE (Mixture of Experts) architecture, featuring robust code generation capabilities and excellent performance in general agent tasks. The model boasts a total of 1 trillion parameters, with 32 billion active parameters. Kimi K2 was officially launched on August 1, initially operating at 10 tokens per second. After optimization, it has now been upgraded to 40 tokens per second and further enhanced to achieve a stable output of 60 tokens per second.
To attract more users, the Kimi K2 Turbo Edition is currently offered at a 50% discount. The input cost for the model is divided into two scenarios: when the cache is hit, the input price is 2 RMB per million tokens, and when the cache is missed, it is 8 RMB. The output price is set at 32 RMB. However, it's important to note that this promotional offer will end on September 1, after which prices will revert to their original rates.
Kimi K2 not only excels in speed but also demonstrates formidable technical prowess. Its strength has earned it widespread recognition and praise from users, particularly in code generation and complex agent tasks. Moonshot AI has stated that it will continue to optimize the model's performance, striving to provide users with more efficient and precise AI services.
Key Highlights:
The output speed of the Kimi K2 Turbo Edition AI model has been increased to 60 tokens per second, with peaks reaching 100 tokens per second.
The model is currently available at a 50% discount, with reasonable input and output pricing to attract users.
Kimi K2 performs exceptionally well in code generation and general agent tasks, receiving high praise from users, with further optimizations planned for the future.