Skip to content

🇨🇳 SuperQode v0.2.22: Kimi K3 Support

Choose a tag to compare

@Shashikant86 Shashikant86 released this 18 Jul 21:24

SuperQode v0.2.22

This release adds first-class support for Moonshot AI’s Kimi K3, including a dedicated long-horizonvcoding harness, provider integration, model metadata, reasoning support, and comprehensivevdocumentation.

Highlights

  • Added Kimi K3 through Moonshot’s first-party global API:
    • Provider: moonshot
    • Model: kimi-k3
    • Endpoint: https://api.moonshot.ai/v1
  • Added MOONSHOT_API_KEY and KIMI_API_KEY credential support.
  • Added current Kimi models to the SuperQode catalog:
    • kimi-k3
    • kimi-k2.7-code-highspeed
    • kimi-k2.7-code
    • kimi-k2.6
  • Added the kimi-k3-coding harness preset with:
    • 1M-token context window
    • Maximum reasoning
    • Parallel tool execution
    • Cache-friendly extended session history
    • Kimi coding-model fallbacks
  • Preserved K3 reasoning_content across multi-turn tool calls.
  • Added K3-specific request shaping:
    • Maps reasoning to the currently supported max level
    • Removes incompatible sampling overrides
    • Converts max_tokens to max_completion_tokens
    • Supports streaming and non-streaming requests
  • Added comprehensive Kimi documentation covering API routes, Kimi Code subscriptions, ACP, pricing,
    vision, tools, caching, structured output, limitations, and troubleshooting.

Getting started

export MOONSHOT_API_KEY=your-api-key

superqode --harness kimi-k3-coding \
  -p "Review this repository and fix the failing tests"

To create a project-owned harness:

superqode harness init kimi-project \
  --template kimi-k3-coding \
  --output harness.yaml

superqode --harness harness.yaml \
  -p "Implement the requested feature"

You can also connect directly:

superqode connect byok moonshot kimi-k3