Kimi Releases K2.7-Code: Open-Source Model Boosts Coding Efficiency
AI company Kimi has launched Kimi K2.7-Code, an open-source model based on Kimi K2.6. It features improved long-horizon coding task completion and a 30% reduction in thinking-token usage. The model utilizes a Mixture-of-Experts architecture with 1 trillion total parameters and a 256K context length, aiming to enhance complex software engineering workflows.
Key points
- Kimi has released Kimi K2.7-Code, an open-source coding-focused AI model built upon Kimi K2.6.
- The new model reportedly improves task completion for real-world long-horizon coding and reduces thinking-token usage by approximately 30% compared to its predecessor.
- Kimi K2.7-Code features a Mixture-of-Experts architecture with 1 trillion total parameters and a 256K context length.
- Evaluations show Kimi K2.7-Code achieving scores of 62.0 on the Kimi Code Bench v2 and 76.0 on MCP Atlas.
- While Kimi K2.7-Code demonstrates improvements, performance on benchmarks like Kimi Code Bench v2 and MLS Bench Lite is still lower than models such as GPT-4.5 and Claude Opus.
AI company Kimi has introduced Kimi K2.7-Code, an enhanced open-source artificial intelligence model specifically designed for coding tasks. Built upon the Kimi K2.6 architecture, the new model promises significant advancements in handling complex, long-horizon coding projects and completing intricate software engineering workflows.
Key improvements in Kimi K2.7-Code include greater token efficiency, with a reported reduction of approximately 30% in the usage of thinking tokens compared to Kimi K2.6. This optimization aims to make complex coding processes more streamlined and less computationally intensive. The model employs a Mixture-of-Experts (MoE) architecture, boasting 1 trillion total parameters and 32 billion activated parameters, alongside an extensive context length of 256,000 tokens.
Initial benchmark results indicate competitive performance in certain areas. For instance, Kimi K2.7-Code scored 62.0 on the Kimi Code Bench v2 and 76.0 on the MCP Atlas benchmark. However, when compared to leading proprietary models like GPT-4.5 and Claude Opus, Kimi K2.7-Code's scores on coding and agentic benchmarks remain lower, suggesting room for further development in the rapidly evolving field of AI coding assistants.
Sources
The WireByte editorial team synthesises technology news from multiple primary sources, verifies the facts, and links every source. Articles are produced with AI assistance and reviewed under our editorial policy.