Moonshot AI's Kimi K2, Now on Together AI, Achieves 65.8% on SWE-bench Verified Coding Tasks

Beijing – Moonshot AI's newly open-sourced Kimi K2 large language model is rapidly gaining traction for its advanced coding and agentic capabilities, now made accessible on platforms like Together AI. This integration empowers developers to engage in "vibe coding" and build sophisticated applications such as subscription trackers, as recently highlighted by a tweet from "AK" stating, "kimi k2 + @togethercompute vibe coding in anycoder build a subscription tracker." The model's release marks a significant development in the open-source AI landscape.

Kimi K2 is a state-of-the-art Mixture-of-Experts (MoE) model boasting 1 trillion total parameters with 32 billion activated per inference. Developed by the Chinese startup Moonshot AI, it is specifically optimized for agentic intelligence, excelling in tool use, reasoning, and autonomous problem-solving. Its performance in coding tasks is particularly noteworthy, achieving a 65.8% single-attempt accuracy on the challenging SWE-bench Verified benchmark.

Together AI has made the Kimi-K2-Instruct variant available on its frontier AI cloud, offering developers high performance, reliability, and scalability. This collaboration provides a cost-effective alternative to proprietary models, with Kimi K2 priced significantly lower than competitors like Anthropic's Claude Opus 4 and OpenAI's GPT-4.1. The model's design allows for seamless integration into existing workflows, supporting complex multi-step tasks and autonomous code generation.

Moonshot AI, founded in March 2023, has quickly emerged as a major player in the AI sector, securing approximately $1.27 billion in funding and reaching a $3.3 billion valuation with backing from major investors including Alibaba and Tencent. The company's strategic decision to open-source Kimi K2 aims to accelerate its adoption and foster innovation within the broader AI community. This move positions Kimi K2 as a direct competitor to leading proprietary and open-source models globally.

The widespread availability and impressive capabilities of Kimi K2 are expected to drive further advancements in AI-assisted development and agentic applications. Its strong performance in real-world coding scenarios and cost-effectiveness make it an attractive option for developers and enterprises seeking to leverage cutting-edge AI for complex problem-solving and automation. The model's open-source nature also encourages collaborative development and customization across diverse use cases.