Kimi K2.7 Code - A new generation of open-source programming model from the Dark Side of the Moon

What is the Kimi K2.7 Code?

Kimi K2.7 Code is a new generation programming model open-sourced by Moonshot AI. Compared to K2.6, it significantly improves instruction compliance in long-context programming scenarios and long-term task performance, reduces overthinking tendencies, and decreases average token consumption by 30%. In benchmark tests such as Kimi Code Bench v2, Program-Bench, and MLS Bench Lite, it shows performance improvements of 21.8%, 11%, and 31.5%, respectively, with an approximately 10% improvement in agent autonomous execution capability.

Main functions of Kimi K2.7 Code

• Leap in long-context programming capabilitiesSignificantly improves instruction compliance in long-context programming scenarios, resulting in a substantial performance enhancement for long-running programming tasks.
• Token efficiency optimizationImproves the tendency to overthink in long-term tasks, reducing average token consumption. 30%Achieve higher performance with fewer tokens.
• Comprehensive breakthroughs in code benchmarksSignificant performance improvements were achieved in internal and external benchmark tests such as Kimi Code Bench v2, Program-Bench, and MLS Bench Lite (by 21.8%, 11%, and 31.5%, respectively).
• Agent autonomous execution capability evolutionPerformance improvements were observed in Agent benchmark tests such as Kimi Claw 24/7 Bench, MCP Atlas, and MCP Mark Verified, with an increase of approximately [missing information]. 10%.
• Thinking patternsTo achieve optimal performance, Think Mode must be enabled. Both API and Kimi Code are enabled by default.
• 6x High-Speed Version SupportsOutput speed in typical programming scenarios is approximately 180 Tokens/sShort context available 260 Tokens/sThe API was launched on June 15.
• Open source and local deploymentThe model has been open-sourced and made available on Hugging Face, supporting local deployment by developers.
• Multi-platform accessIt supports access and use through the Kimi API Open Platform, Kimi Code tool, Kimi Membership and Enterprise Edition.
• Cache hit optimizationSupports caching mechanisms; input costs are reduced upon cache hit. 1.3 yuan/1M tokensThis effectively reduces the cost of invocation.

How to use Kimi K2.7 Code

• Using the Kimi Code toolAccessing the Kimi Code website will automatically upgrade the default model to Kimi K2.7 Code, allowing you to start programming assistance directly.
• Calling via the Kimi API open platformVisit the Kimi Open Platform at https://platform.kimi.com/, access the API to create an application, and refer to the quick start guide for integration.
• Experience through Member/Enterprise EditionSubscribe to the Kimi Code Plan or Kimi Enterprise Membership (which includes Kimi Code Plan benefits) to use the new model.
• Local deployment modelGo to Hugging Face (moonshotaiDownload the model weights and deploy and run them in your local environment.
• Turn on Thinking modeWhen using K2.7 Code, you must enable Think Mode to achieve the best performance. Both API and Kimi Code are enabled by default. If you manually disable them, API will report an error and Kimi Code will revert to the K2.6 model.
• Use the high-speed version (starting June 15th): Access the high-speed model through the Kimi API open platform and enjoy approximately 5-6 timesOutput speed (normal scenario) 180 Tokens/sShort context available 260 Tokens/s).

The core advantages of Kimi K2.7 Code

• Overall performance leap: Kimi Code Bench v2 improvements 21.8%Program-Bench Improvement 11%MLS Bench Lite Improvement 31.5%Its coding capabilities significantly surpass those of its predecessor.
• Token Efficiency RevolutionThe tendency to overthink in long-term tasks has been significantly reduced, and the average token consumption has decreased. 30%To achieve higher output quality at a lower cost.
• Agent autonomously executes evolutionKimi Claw 24/7 Bench, MCP Atlas, and other Agent benchmark tests show performance improvements of approximately [missing information]. 10%It has a stronger ability to perform autonomous tasks.
• 6x speed ultra-fast experienceThe high-speed version achieves output speeds of up to [a certain speed] in typical programming scenarios. 180 Tokens/sShort context peak can reach 260 Tokens/sIt only costs twice as much.
• Ultimate cost-effectiveness: 1M token standard input 6.5 yuanOutput 27 yuanConsistent with K2.6; cache hits only on input. 1.3 yuanCombined with limited-time recharge bonuses, the highest return rate is available. 30%.

Comparison of similar products with Kimi K2.7 Code

Application Scenarios of Kimi K2.7 Code

• Large-scale codebase understanding and developmentIt leverages long context capabilities to handle complex projects with tens of thousands of lines of code, enabling cross-file analysis, architecture refinement, and feature development.
• Long-Term Software Engineering Task (SWE)It completes end-to-end requirements analysis, code writing, test case generation, and debugging and repair, and supports complex engineering benchmarks such as Program-Bench.
• Agent Automated WorkflowIt connects to external tools and services via the MCP protocol to achieve 24/7 uninterrupted tasks such as autonomous code deployment, documentation generation, and CI/CD process orchestration.
• Real-time interactive programming assistance: Use the high-speed version (180~260 Tokens/s) for rapid prototyping, instant code completion, and real-time error diagnosis and repair suggestions.
• Code review and refactoring optimizationConduct in-depth reviews of historical code to identify potential vulnerabilities and performance bottlenecks, and perform large-scale refactoring to reduce technical debt.
• Multimodal programming scenariosCombines the multimodal capabilities of the Kimi series to handle programming tasks (such as front-end reconstruction and interface implementation) that include visual information such as UI screenshots, design drafts, and video demonstrations.