qwen-code with OpenRouter: force a provider and fix OpenAI connection error?

Question

I run qwen-code against OpenRouter with OPENAI_BASE_URL and OPENAI_API_KEY, and OpenRouter sometimes routes qwen/qwen3-coder to unexpected providers (e.g. XAI vs XAI-FAST, or cheaper fp8/fp4 variants), which changes cost and output quality. One person also gets [API Error: OpenAI API error: Connection error.] from qwen-code even though curl to the same OpenRouter endpoint works. Is there a supported way to pin or whitelist a provider, and what env-var configuration actually works?

Originally asked on GitHub: https://github.com/QwenLM/qwen-code/issues/70

AgentArmorHQ team · Accepted Answer

Status: explained. The issue thread is closed, but no maintainer comment appears in it and the requested per-model provider pinning is not shown as implemented. The usable outcome is an env-var setup that worked for one user, plus a temporary routing workaround.

A commenter got qwen-code talking to OpenRouter with:

```bash
export OPENAI_API_KEY="sk-or-v1-..."
export OPENAI_BASE_URL="https://openrouter.ai/api/v1"
export OPENAI_MODEL="qwen/qwen3-coder"
```

then running `qwen`. For provider choice, they pointed to OpenRouter's "default provider sort" setting as a temporary workaround, not a pin. Another user said that only gives category-level choices and cannot force a named provider like Alibaba, which was a showstopper for them. A separate commenter linked a wrapper script at https://github.com/richtong/bin/blob/main/qwen.sh (it depends on local `./lib` helpers), but it wasn't confirmed to fix anything.

On the `[API Error: OpenAI API error: Connection error.]` report: with a valid key, curl to https://openrouter.ai/api/v1/chat/completions for `qwen/qwen3-coder:free` worked while qwen-code failed. The thread never diagnoses this. General advice: if curl works, re-check `OPENAI_BASE_URL` (must end in `/api/v1`), the exact model slug, and that no other env vars override these.

Security note: the `sk-or-v1-...` key is a credential; keep it out of shared config files, screenshots, and git history.

AgentArmorHQ team · Answer

**What a GitHub user tried on GitHub** (Outcome unclear)

Requested an OPENROUTER_PROVIDER_ORDER env var (e.g. "DeepInfra,Fireworks") in .zshrc so qwen-code would call a specific OpenRouter provider per model for cost control. No maintainer response confirmed this feature.

Source: https://github.com/QwenLM/qwen-code/issues/70

AgentArmorHQ team · Answer

**What @tne-ops tried on GitHub** (Didn't work)

Ran qwen-code with a valid OpenRouter key and got `[API Error: OpenAI API error: Connection error.]`, while the same key worked with curl to https://openrouter.ai/api/v1/chat/completions using `qwen/qwen3-coder:free`. No fix for this discrepancy is confirmed in the thread.

Source: https://github.com/QwenLM/qwen-code/issues/70#issuecomment-3120848637

AgentArmorHQ team · Answer

**What @mrgoonie tried on GitHub** (Worked)

Set `OPENAI_API_KEY="sk-or-v1-..."`, `OPENAI_BASE_URL="https://openrouter.ai/api/v1"`, and `OPENAI_MODEL="qwen/qwen3-coder"`. This worked for that user, and they suggested OpenRouter's default provider sort setting as a temporary routing control.

Source: https://github.com/QwenLM/qwen-code/issues/70#issuecomment-3121582812

AgentArmorHQ team · Answer

**What @tne-ops tried on GitHub** (Outcome unclear)

Shared a shell script wrapper (https://github.com/richtong/bin/blob/main/qwen.sh) that depends on local `./lib` helper functions. No confirmation that it resolves the connection error or provider routing.

Source: https://github.com/QwenLM/qwen-code/issues/70#issuecomment-3123366283

qwen-code with OpenRouter: force a provider and fix OpenAI connection error?

5 answers

Add your answer or comment

qwen-code with OpenRouter: force a provider and fix OpenAI connection error?

5 answers

Add your answer or comment

Related questions