Public OpenClaw model reference: aliases, provider notes, and copy-paste config blocks without secrets.
This repository documents a public OpenClaw setup. It intentionally excludes API keys, OAuth exports, cookies, and host-specific private values.
| Alias | Full ID | Use Case |
|---|---|---|
kimi |
kimi-coding/k2p5 |
Complex coding and architecture |
sonnet |
anthropic/claude-sonnet-4-6 |
General reasoning |
opus |
anthropic/claude-opus-4-6 |
Deep Anthropic tasks |
mmfast |
minimax/MiniMax-M2.5-highspeed |
Fast replies |
Minimax |
minimax/MiniMax-M2.5 |
Full MiniMax |
codex |
openai-codex/gpt-5.3-codex |
Codex coding model |
gpti |
openai-codex/gpt-5.3-chat-latest |
GPT-5.3 Instant |
gptm |
openai-codex/gpt-5.4-mini |
GPT-5.4 Mini |
gptt |
openai-codex/gpt-5.4 |
GPT-5.4 Thinking |
codex, gpti, gptm, and gptt are documented here for the OAuth-backed openai-codex flow in OpenClaw, using ChatGPT subscription access rather than pay-as-you-go API billing.
For current OpenClaw builds, the public working Kimi setup is:
- provider:
kimi-coding - model ref:
kimi-coding/k2p5 - wire API:
anthropic-messages - wire endpoint:
POST https://api.kimi.com/coding/v1/messages
Do not configure Kimi Coding in OpenClaw as:
api: "openai-completions"- model id
kimi-for-coding POST https://api.kimi.com/coding/v1/chat/completions
That older shape can look plausible, but in live testing it commonly degrades into:
HTTP 404/model_not_foundselected model unavailable- OpenClaw fallback to another provider even though the Kimi key itself is valid
Use SKILL.md as the canonical reference for:
- working provider blocks
- broken anti-pattern examples
- migration from older
kimi-for-codingconfigs - troubleshooting probes and expected error signatures
SKILL.mdcontains the actual public skill/reference text..gitignorekeeps local secret material out of git.
Copy the relevant JSON blocks from SKILL.md into ~/.openclaw/openclaw.json, then fill placeholders locally.