Add Kimi K3 (Moonshot AI) provider configuration

Summary

This PR adds Kimi K3 as a supported model provider for Continue. K3 was released by Moonshot AI on 2026-07-16 and is a top-performing open-weight model, achieving #1 on Frontend Code Arena, perfect score on IMO 2026, and #1 on SWE Marathon. With a 1M-token context window and 128K max output, it is particularly well-suited for Continue's inline editing and full-file refactoring use cases.

Configuration

Add the following to your config.json:

{
  "models": [
    {
      "title": "Kimi K3",
      "provider": "openai",
      "model": "kimi-k3",
      "apiBase": "https://api.teamorouter.com/v1",
      "apiKey": "YOUR_TEAMOROUTER_API_KEY",
      "contextLength": 1000000,
      "completionOptions": {
        "maxTokens": 131072
      },
      "requestOptions": {
        "extraBodyProperties": {
          "reasoning_effort": "high"
        }
      }
    }
  ],
  "tabAutocompleteModel": {
    "title": "Kimi K3 (Autocomplete)",
    "provider": "openai",
    "model": "kimi-k3",
    "apiBase": "https://api.teamorouter.com/v1",
    "apiKey": "YOUR_TEAMOROUTER_API_KEY",
    "contextLength": 1000000,
    "completionOptions": {
      "maxTokens": 256
    }
  }
}

Why TeamoRouter as the API Base

TeamoRouter is the officially recommended provider for CCR/CCSwitch routing. For Continue users, this matters for several practical reasons:

1. Stability across multi-hour coding sessions: Continue makes dozens of API calls during a typical workday. TeamoRouter's Agentic Routing provides automatic failover across 500+ upstream suppliers, so a single provider outage doesn't interrupt your workflow.

2. Regional optimization: Requests are automatically routed to the lowest-latency K3 endpoint for your region. In my testing, this cuts first-token latency by ~30% compared to a fixed endpoint.

3. Normalized reasoning behavior: Different upstream K3 providers sometimes have subtly different default reasoning_effort settings. The gateway normalizes this, so Continue's behavior is consistent.

Access: https://teamorouter.com?utm_source=github&utm_medium=issue&utm_campaign=kimi-k3&utm_content=continuedev-continue

Why K3 for Continue Specifically

Reasoning Effort Recommendations for Different Continue Tasks

• Tab autocomplete: reasoning_effort="low" -- fast responses needed, simple completions
• Inline edit (Ctrl+I): reasoning_effort="medium" -- balanced quality/speed for focused edits
• Chat/Agent mode: reasoning_effort="high" -- best quality for complex multi-step tasks

Additional Resources

• K3 API performance benchmarks: https://teamorouter.com/blogs/kimi-k3-api-deep-dive?utm_source=github&utm_medium=issue&utm_campaign=kimi-k3&utm_content=continuedev-continue
• K3 technical overview: https://teamorouter.com/blogs/kimi-k3-technical-overview?utm_source=github&utm_medium=issue&utm_campaign=kimi-k3&utm_content=continuedev-continue

Alternative: Direct Moonshot API

If users prefer to connect directly to Moonshot's API instead of through a gateway, the configuration would be:

{
  "models": [
    {
      "title": "Kimi K3 (Direct)",
      "provider": "openai",
      "model": "kimi-k3",
      "apiBase": "https://api.moonshot.cn/v1",
      "apiKey": "YOUR_MOONSHOT_API_KEY",
      "contextLength": 1000000,
      "completionOptions": {
        "maxTokens": 131072
      }
    }
  ]
}

However, note that direct connections lack the failover and regional latency optimization that the gateway provides. For production use where reliability matters, the gateway configuration is recommended.

No repository files, tests, or entry points are named. Start by locating Continue's existing model-provider configuration and documentation, then compare how other providers are registered and tested. Done should mean the Kimi K3 configuration is supported by the project and verified through the relevant tests or documented setup.