Moonshot AI ships Kimi K2.7-Code targeting token efficiency in agentic coding

Moonshot AI shipped Kimi K2.7-Code on June 12, 2026. The model is the fifth major release in the Kimi series in under a year. It is open-source under a Modified MIT license on Hugging Face and accessible via the Kimi API and Kimi Code CLI. Moonshot says K2.7-Code improves 21.8 percent on Kimi Code Bench v2 over K2.6 and cuts reasoning token usage by 30 percent. The model uses a Mixture-of-Experts architecture with 1 trillion total parameters and 32 billion active per token along with a 256K-token context window. Thinking mode is forced on and cannot be disabled. The company reported gains on its internal benchmarks of 11.0 percent on Program Bench and 31.5 percent on MLS Bench Lite versus K2.6. All benchmarks are Moonshot proprietary. No independent third-party results on SWE-bench Verified, LiveCodeBench, or GPQA Diamond were available as of the release date. K2.7-Code scored 81.1 on MCP Mark Verified and supports multimodal input including image and video. Mitch Ashley, VP and practice lead for software lifecycle engineering and AI-native software engineering at The Futurum Group, said token efficiency is a transitory challenge in agentic coding. He noted that such gains get absorbed into base capabilities over release cycles. Kimi Code, the company's terminal-first coding agent, starts at 19 dollars per month. API pricing is 0.95 dollars per million input tokens and 4.00 dollars per million output tokens. Weights are on Hugging Face with support for vLLM, SGLang, or KTransformers and reuse of K2.6 deployment patterns.