Back to rankings

GitHub Copilot: Kimi K3 Model Now Available

GitHub CopilotView original changelog

GitHub Copilot now supports Kimi K3, Moonshot AI's open-weight, 2.8-trillion-parameter model hosted on Fireworks AI, bringing frontier-level agentic coding performance at cost-effective pricing. The model ships with a 1-million-token context window and native vision capabilities, priced at $3 per 1M input tokens, $15 per 1M output tokens, and $0.30 per 1M cached input tokens. It rolls out gradually to Copilot Pro, Pro+, Max, Business, and Enterprise plans across VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the Copilot CLI, cloud agent, and app. Business and Enterprise organizations must have an administrator enable it via policy, since it ships off by default.

Key Takeaways

  • Kimi K3 is a 2.8-trillion-parameter open-weight model from Moonshot AI, hosted on Fireworks AI, now integrated into GitHub Copilot.
  • The model ships with a 1-million-token context window and native vision capabilities, enabling it to process large codebases and multimodal input in a single request.
  • Pricing is highly competitive at $3 per 1M input tokens and $15 per 1M output tokens, with a steep discount to $0.30 per 1M for cached input tokens.
  • The rollout spans nearly every Copilot surface, including VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the Copilot CLI, cloud agent, and app.
  • It's available on Pro, Pro+, Max, Business, and Enterprise plans, though it ships off by default for Business and Enterprise until an admin enables it.
  • Kimi K3 follows Kimi K2.7 Code as the second Moonshot AI model added to Copilot, reinforcing GitHub's strategy of diversifying its model lineup beyond OpenAI, Anthropic, and Google.

Kimi K3 Joins GitHub Copilot's Model Lineup

GitHub Copilot has added support for Kimi K3, an open-weight model developed by Moonshot AI and hosted on Fireworks AI. According to GitHub's announcement, the model demonstrates "frontier-level abilities on agentic coding with highly cost-effective pricing," positioning it as a budget-friendly alternative to the proprietary frontier models already available inside Copilot.

Model Capabilities

Kimi K3 is built on a 2.8-trillion-parameter architecture, making it one of the largest open-weight models available in Copilot's lineup. It includes native multimodal vision capabilities and a 1-million-token context window, letting it reason over large codebases, long conversation histories, or extensive documentation in a single request without needing the input broken into smaller chunks.

Pricing

Kimi K3 is billed under GitHub Copilot's usage-based pricing model: $3 per million input tokens, $15 per million output tokens, and a discounted $0.30 per million for cached input tokens. The cached-token rate stands out for agentic workflows that repeatedly reference the same context, such as long-running coding sessions or multi-step agent tasks, since it substantially reduces the effective cost of extended sessions.

Availability

The model is rolling out gradually across nearly every Copilot surface, including VS Code, Visual Studio, JetBrains, Xcode, Eclipse, the Copilot CLI, the Copilot cloud agent, the standalone GitHub Copilot app, github.com, and GitHub Mobile for iOS and Android. It's available to Copilot Pro, Pro+, Max, Business, and Enterprise subscribers.

For Business and Enterprise customers, Kimi K3 is off by default. Administrators need to review the model's security and compliance profile and explicitly enable it in Copilot's model policy settings before members of their organization can select it. GitHub says the rollout will continue to be monitored as it reaches more users.

Why It Matters

Kimi K3 extends Copilot's growing roster of third-party and open-weight models, following the earlier addition of Kimi K2.7 Code, giving developers another option to balance cost against capability for agentic coding tasks. Its low per-token pricing, especially the steep discount on cached input, could make it an attractive default for teams running frequent or long-running agent sessions where costs add up quickly under usage-based billing.