Kimi K2.7 Code: The first open-weight model in GitHub Copilot

By GitHub

Categories: Product, Tools

Summary

GitHub Copilot now includes Kimi K2.7 Code, the first open-weight coding model in its lineup, priced at just $0.95 per million input tokens—95% cheaper than frontier models while matching their agentic coding accuracy and delivering 95% cache hit efficiency.

Key Takeaways

  1. Kimi K2.7 Code costs 95 AI credits ($0.95) per million input tokens and 400 credits per million output tokens, positioning it among the cheapest models in GitHub Copilot while matching frontier model performance.
  2. Open-weight models like Kimi enable customization and fine-tuning for organizations, allowing teams to adapt the model to domain-specific coding tasks without vendor lock-in.
  3. Kimi demonstrated strong reasoning ability by identifying assumptions and asking 8 clarifying questions when planning a full product requirements document, then successfully completed all implementation tasks with minimal follow-up.
  4. The model achieves 95% cache hit rate and is among the fastest for time-to-first-token and time-to-completion metrics, making it efficient for both rapid iteration and token cost optimization.
  5. Kimi is disabled by default for organizations; developers should check the Models tab in Copilot CLI to verify which models are enabled for their account before attempting to use it.

Related topics

Transcript Excerpt

Today, GitHub is announcing that the Kimi K2.7 Code model has been added to the GitHub Copilot model lineup. Now, what is this Kimi model and what do you need to know about it? Let's take a look. Kimi is a model from a company called Moonshot AI, and K2.7 Code is the most recent iteration of their coding focused agentic model. This model is what's known as an open weights model. This means that the parameters that were used to train Kimi are made available to everyone. So this model can be altered or fine-tuned. And this makes Kimi the first open weights model to be added to the GitHub Copilot model lineup. Kimi models are hosted on US-based Microsoft AI Foundry infrastructure and managed by GitHub and Microsoft. To use Kimi K2.7 Code, just find it in your model picker. Now, note that Kimi…

More from GitHub