Moonshot AI, backed by Alibaba, has released Kimi K2, its latest open‑source large language model built with a mixture‑of‑experts architecture. The model activates 32 billion parameters per inference from a total of 1 trillion parameters. Early benchmark reports indicate Kimi K2 surpasses GPT‑4.1 and Anthropic’s Claude Opus on several coding and agentic tasks. This release strengthens China’s push to regain AI leadership through open‑source innovation.

Kimi K2’s pricing strategy undercuts major players. It offers $0.15 per million input tokens and $2.50 per million output tokens, a substantial cost reduction compared to OpenAI and Anthropic’s flagship models. Developers can also access Kimi K2 for free via the Moonshot app or web interface, and both base and instruct versions of the model are open‑sourced on GitHub. This dual strategy combines accessibility with enterprise‑grade capability.

For developers and enterprise teams, Kimi K2 provides a new contender for advanced coding and agentic workflows without the heavy costs of proprietary models. It introduces competitive performance for AI‑driven applications like code generation, software agents, and domain‑specific AI tasks. By integrating trillion‑scale models into the open‑source landscape, Moonshot AI positions itself as a key challenger in the escalating global LLM competition.