Pick Qwen by the workload and endpoint. Model availability was checked on October 14, 2026 against the haimaker catalog and Ollama’s Qwen3.8 listing. Cloud Qwen3.7 IDs and local Qwen3.8 tags are different products; do not substitute one name for the other.
The quick answer
| Model | Endpoint/model ID | Workload to test |
|---|---|---|
| Qwen3.7 Plus | qwen/qwen3.7-plus through haimaker | Reasoning and general agent work |
| Qwen3.7 Max | qwen/qwen3.7-max through haimaker | Harder cloud tasks |
| Qwen3.7 Flash | qwen/qwen3.7-flash through haimaker | Lighter, high-volume work |
| Qwen3 Coder | qwen/qwen3-coder through haimaker | Coding-focused baseline |
| Qwen3.8 27B | qwen3.8:27b through local Ollama | Private coding on suitable hardware |
Qwen3 Coder — the coding baseline
Keep it in the comparison when coding is the main workload. Run the same edit and test sequence on it and a current cloud candidate, then compare correctness, tool errors, and total cost.
Qwen3.7 Plus / Max — cloud reasoning candidates
These IDs are available in the current haimaker catalog. Try them for longer planning or debugging tasks, with the same tools and context. Availability does not establish which one is better on your repository.
Qwen3.7 Flash — lighter cloud work
Test Flash on summaries, classification, and routine agent tasks. Inspect input and output rates separately; retries can erase a low token price.
Qwen3.8 27B — local coding and tools
Use ollama pull qwen3.8:27b on suitable hardware. The default download is about 18GB before context and runtime buffers. A 24GB GPU or 32GB+ unified-memory machine is a starting point for a bounded context, not a guarantee that the maximum window fits.
Older models and existing setups
A working older setup can remain a useful baseline. Re-evaluate it against current candidates on the same task before changing the default. Do not copy old benchmark scores, cost ratios, or hardware figures onto newer versions.
Setup in OpenClaw
The fastest path to Qwen in OpenClaw is through haimaker.ai. Qwen models are available alongside Claude, GPT, Gemini, and dozens of other models through a single API key. No separate Alibaba Cloud account needed.
1. Get your haimaker.ai API key
Sign up at haimaker.ai and copy your key from the dashboard.
2. Add haimaker as a provider
Open ~/.openclaw/openclaw.json:
{
"models": {
"providers": {
"haimaker": {
"baseUrl": "https://api.haimaker.ai/v1",
"apiKey": "your-haimaker-api-key",
"api": "openai-completions"
}
}
}
}
3. Add Qwen models to the allowlist
{
"agents": {
"defaults": {
"models": {
"haimaker/qwen/qwen3-coder": {},
"haimaker/qwen/qwen3.7-plus": {},
"haimaker/qwen/qwen3.7-flash": {}
}
}
}
}
4. Apply the config
Run openclaw gateway config.apply and switch models with /model during a session.
Want to skip those four steps? npx -y @haimaker/connect --openclaw writes the provider and registers the Qwen models in your allowlist automatically — --model qwen/qwen3-coder defaults you to the coding pick. Setup details live in the connect guide.
What I’d do
Compare Qwen3 Coder with Qwen3.7 Plus on one coding task, and test Qwen3.7 Flash on a routine job. For private local work, compare Qwen3.8 with a local coding baseline. Use the model that finishes correctly at an acceptable total cost, and copy its exact ID from the endpoint you configure.