Moonshot AI Launches Kimi K2, Beating ChatGPT in Coding Tests

Alibaba-backed Moonshot AI released Kimi K2, an open-source model, claiming it outperforms OpenAI’s ChatGPT and Anthropic’s Claude in coding benchmarks while offering lower API pricing than competitors.

The model uses a trillion-parameter mixture-of-experts design with 32 billion active parameters. Moonshot optimized it for code generation, autonomous agent tasks, and tool integration.

Kimi K2 scored 65.8% accuracy on the SWE-bench Verified software-engineering test and 53.7% on LiveCodeBench. GPT-4.1 scored 54.6% on SWE-bench Verified and 44.7% on LiveCodeBench. The model matched or exceeded other leading open-source alternatives in performance.

Moonshot also priced Kimi K2’s API at $0.15 per million input tokens for cache hits and $2.50 per million output tokens. This pricing undercuts both OpenAI and Anthropic’s paid plans.

The company described Kimi K2 as a model that “does not just answer; it acts.” The system can autonomously write and execute code and orchestrate multi-step workflows without human intervention.

Chinese tech firms are increasingly open-sourcing advanced AI models. This strategy aims to build global developer communities and counter U.S. export restrictions on cutting-edge models.

Yang Zhilin, a Tsinghua University graduate, founded Moonshot AI in 2023 with backing from Alibaba. The company initially gained attention for its long-text analysis tools. However, DeepSeek’s low-cost models cut into Moonshot’s user base. 

Kimi K2 represents Moonshot’s attempt to regain market share by combining high coding performance with competitive pricing designed to attract enterprise customers from established providers.

The model’s release comes as AI labs worldwide face challenges with the economics of training larger models. Moonshot’s focus on both performance and cost-effectiveness could influence how enterprises adopt and deploy generative AI systems.