Alibaba-backed Chinese AI startup Moonshot AI has released a new open-source model, Kimi K2. Moonshot AI claims it excels in advanced knowledge, math, programming, and agent tasks.
In the announcement, the startup said that its neural network outperformed Claude Opus 4 on two benchmarks and performed better than OpenAI's GPT-4.1 on several metrics. Additionally, the model offers lower token processing costs:
$0.15 per 1 million input tokens
$2.50 per 1 million output tokens
By comparison, Claude Opus 4 charges $15 for 1 million input tokens (100 times more) and $75 for 1 million output tokens (30 times more). For OpenAI's GPT-4.1, the rates are $2 for 1 million input tokens and $8 for 1 million output tokens.
Kimi K2 is based on the Mixture-of-Experts (MoE) architecture and contains 1 trillion parameters, with 32 billion active at any given time. These active parameters are specialized compute blocks assigned to specific tasks.
Kimi K2 is available for free through the Kimi app and web interface. The company has open-sourced two versions:
Kimi-K2-Base – a base model designed for researchers and developers who need full control over customization
Kimi-K2-Instruct – an instruction-tuned version optimized for universal applications, including chatbots and agent-based AI scripting
Initial reviews of Kimi K2 on English and Chinese social networks are mostly positive, although some users have noted the widespread issue of hallucinations in the AI segment.
Open Source Is Trending
The new model reflects a broader industry trend: the move to open source, which allows both startups and large technology companies to improve efficiency and accelerate the adoption of AI products. This approach involves providing public access to a program’s source code, enabling third-party developers to:
Modify and refine the model architecture
Fix bugs
Scale functionality
Implement their own improvements and customizations for specific tasks
Among prominent Chinese startups, DeepSeek and Tencent offer open-source models.Among American competitors, Meta and, to some extent, Google have also adopted this approach.
At the same time, OpenAI CEO Sam Altman announced that the launch of the company’s own open-source AI model has been postponed indefinitely, citing the need for additional security tests.
