Kimi K2.7 Code
Moonshot AI's coding-focused agentic model with open weights: 1T-parameter MoE, 32B active, 256K context.
Kimi K2.7 Code is a coding model from Moonshot AI built on Kimi K2.6, with code and weights released under a Modified MIT License. Kimi's API charges $0.95 per million input tokens, $4.00 output and $0.19 cached, but stops offering K2 models to new users on October 31, 2026 and plans to retire them by the end of December 2026.
About Kimi K2.7 Code
Moonshot says K2.7 Code improves on Kimi K2.6 at real-world long-horizon coding tasks and at end-to-end task completion across complex software engineering workflows. It is a Mixture-of-Experts model with 1T total and 32B activated parameters, 384 experts with 8 selected per token, a 256K context length and a 400M-parameter MoonViT vision encoder, and it accepts image and video input.
Moonshot's API is OpenAI- and Anthropic-compatible, and Moonshot says the model works best with Kimi Code CLI as its agent framework. Kimi's API bills pay-as-you-go with no subscription plan; a kimi-k2.7-code-highspeed variant, the same model at about 180 output tokens per second and up to 260 in short contexts, costs $1.90 per million input tokens, $8.00 output and $0.38 cached.
Moonshot recommends running it on vLLM, SGLang or KTransformers. Its deployment guide serves the model on a single H200 node with TP8, recommends vLLM 0.19.1 for stable production use and SGLang v0.5.10 or later, and shows KTransformers with SGLang for CPU+GPU heterogeneous inference and KTransformers with LLaMA-Factory for LoRA fine-tuning.
Moonshot's evaluation table scores K2.7 Code above K2.6 on all six coding and agentic benchmarks it lists. Two of them, Kimi Code Bench v2 and Kimi Claw 24/7 Bench, are Moonshot's in-house benchmarks, and Moonshot ran its own models through Kimi Code CLI with thinking on, GPT-5.5 in Codex and Opus 4.8 in Claude Code.
Sources: huggingface.co/moonshotai/Kimi-K2.7-Code, /moonshotai/Kimi-K2.7-Code/blob/main/docs/deploy_guidance.md, platform.kimi.ai/docs/pricing/chat, platform.kimi.ai/docs/guide/kimi-k2-7-code-quickstart and /moonshotai/Kimi-K2.7-Code/blob/main/LICENSE, read 2026-10-27
Kimi K2.7 Code pros and cons
Pros
- The Modified MIT License adds one condition: a commercial product or service with more than 100 million monthly active users or more than 20 million US dollars in monthly revenue must prominently display Kimi K2.7 Code in its user interface.
- On Kimi's API, input and output cost the same as Kimi K2.6, $0.95 and $4.00 per million tokens.
- Moonshot says it uses about 30% fewer thinking tokens than Kimi K2.6.
- Moonshot says the Kimi-K2.5 and K2.6 deployment method can be reused directly, because the architecture is the same.
Cons
- Newly registered users of Kimi's API can no longer use K2 series models, K2.7 Code included, from October 31, 2026, and Kimi plans to retire all K2 series models by the end of December 2026.
- Thinking cannot be switched off: Kimi's guide says K2.7 Code does not support non-thinking mode, and preserve_thinking is enabled by default and can't be disabled.
- Chat with video is experimental and works only on Moonshot's official API, not on third-party vLLM or SGLang deployments.
- Moonshot's own table puts it below GPT-5.5 on all six benchmarks and below Claude Opus 4.8 on five, ahead only on MCP Mark Verified.
Tagged
- Web-based
Kimi K2.7 Code alternatives
The closest alternatives to Kimi K2.7 Code.
- 01→Kimi K3
Moonshot AI's 2.8T-parameter model with native vision and a 1M-token context for long-horizon coding.
- 02→LongCat-2.0
Meituan's open-weights 1.6T-parameter MoE language model for coding and agentic tasks, with 1M-token context.
- 03→DeepSeek-V4-Flash-0731
Official release of DeepSeek-V4-Flash, with stronger agentic skills and open weights under the MIT License.
Related reading
Featured on Vibedonalds
Own Kimi K2.7 Code? Add this badge to your site to show you’re listed — and link back to your profile here.
<a href="https://vibedonalds.com/tools/kimi-k2-7-code" target="_blank" rel="noopener">
<img src="https://vibedonalds.com/badge/featured-on-vibedonalds.svg" alt="Kimi K2.7 Code — Featured on Vibedonalds" width="240" height="60" loading="lazy" />
</a>Frequently asked questions
- What is Kimi K2.7 Code?
- Kimi K2.7 Code is a moonshot AI's coding-focused agentic model with open weights: 1T-parameter MoE, 32B active, 256K context.
- What are the alternatives to Kimi K2.7 Code?
- The closest alternatives to Kimi K2.7 Code are Kimi K3, LongCat-2.0, DeepSeek-V4-Flash-0731.
- What are the downsides of Kimi K2.7 Code?
- Newly registered users of Kimi's API can no longer use K2 series models, K2.7 Code included, from October 31, 2026, and Kimi plans to retire all K2 series models by the end of December 2026. Thinking cannot be switched off: Kimi's guide says K2.7 Code does not support non-thinking mode, and preserve_thinking is enabled by default and can't be disabled. Chat with video is experimental and works only on Moonshot's official API, not on third-party vLLM or SGLang deployments. Moonshot's own table puts it below GPT-5.5 on all six benchmarks and below Claude Opus 4.8 on five, ahead only on MCP Mark Verified.
- Is Kimi K2.7 Code open source?
- Moonshot releases both the code repository and the model weights under the Modified MIT License. Its only modification: a commercial product or service with more than 100 million monthly active users or more than 20 million US dollars in monthly revenue must prominently display Kimi K2.7 Code on its user interface.
- How much does the Kimi K2.7 Code API cost?
- Kimi's API lists kimi-k2.7-code at $0.95 per million input tokens, $4.00 per million output tokens and $0.19 per million cached input tokens, and kimi-k2.7-code-highspeed at $1.90, $8.00 and $0.38. Billing is pay-as-you-go with no subscription plan, and prices exclude applicable taxes.
- Is the Kimi K2.7 Code API being retired?
- Kimi says K2 series models will no longer be available to newly registered users from October 31, 2026, while existing users are not affected, and that it plans to retire all K2 series models by the end of December 2026. The weights are published on Hugging Face under the Modified MIT License.
- How do I run Kimi K2.7 Code myself?
- Moonshot recommends vLLM, SGLang or KTransformers and says the Kimi-K2.5 and K2.6 deployment method can be reused. Its guide serves the model on one H200 node with TP8, recommends vLLM 0.19.1 for stable production or SGLang v0.5.10 and later, and needs transformers version 4.57.1 or later, below 5.0.0.
- How does Kimi K2.7 Code compare with GPT-5.5 and Claude Opus 4.8?
- In Moonshot's own evaluation table it scores below GPT-5.5 on all six coding and agentic benchmarks listed and below Claude Opus 4.8 on five, beating Opus 4.8 on MCP Mark Verified, 81.1 to 76.4. Moonshot tested its models in Kimi Code CLI, GPT-5.5 in Codex and Opus 4.8 in Claude Code.