Qwen Code shows far fewer tokens than Bailian bills — how do I stop charges and get a refund?
Qwen Code with Alibaba Cloud Bailian shows a small token count in the CLI while the Bailian console bills many times more (e.g., 40K shown vs 200K billed). Users are seeing unexpected charges and want to know how to stop the spending and dispute the costs. What steps actually worked in this thread?
Originally asked on GitHub: https://github.com/QwenLM/qwen-code/issues/83
6 answers
Status: explained. The GitHub issue was closed as completed, but the thread contains no code fix or fixed version. The Qwen Code homepage carries a warning (quoted in the thread) stating that Qwen Code may issue multiple API calls per cycle, resulting in higher token usage, and that the team is working on API efficiency. Bailian support reportedly attributed the overspend to the prompts sent by qwen-code, but that is a secondhand account from the thread.
What worked in the thread:
- Delete your API key first so charges stop accumulating (comment by a user who filed a dispute).
- File a support ticket with Alibaba Cloud/Bailian. The thread reporter (issue body) contacted their business contact and Alibaba Cloud technical support, and after troubleshooting, received a voucher refund (代金券). Another commenter got the same approach rolling: after filing a ticket, a business representative called and said they would consult engineers.
- If first-line support is unhelpful, ask for evidence of the actual input content sent to the API, and ask to be escalated to a representative with higher permissions (comment 4).
Additional finding: one commenter tested Kilo Code (not qwen-code) with Qwen3-coder for about 3 hours. The CLI showed 23.3K tokens, but the Bailian platform later showed 224.1K tokens across 10 API requests. The commenter suspected Qwen3-coder does not support prompt caching, so each request bills the full context. This is a hypothesis, not confirmed, and it suggests the billing discrepancy is not limited to the qwen-code CLI.
Did this work for you?What a GitHub user tried on GitHub (Worked)
The reporter fixed three faults (20M+ tokens) then compared a similar fix: Cline consumed about 600K tokens while Qwen Code consumed 8M. After contacting their business contact and Alibaba Cloud technical support, they received a voucher refund.
Source: https://github.com/QwenLM/qwen-code/issues/83 (opens external site)
Did this work for you?What @Amoyens1s tried on GitHub (Outcome unclear)
Filed a support ticket; a business representative called and said they would consult engineers. The commenter advised asking support for the actual input content and escalating to a higher-permission representative if first-line support deflects. No final refund is stated.
Source: https://github.com/QwenLM/qwen-code/issues/83#issuecomment-3111891380 (opens external site)
Did this work for you?What @han0420 tried on GitHub (Outcome unclear)
Advised deleting the API key first and then filing a support ticket to recover charges, quoting the homepage warning about multiple API calls per cycle. No outcome for this advice is reported.
Source: https://github.com/QwenLM/qwen-code/issues/83#issuecomment-3112140549 (opens external site)
Did this work for you?What @beagle4ce tried on GitHub (Partly worked)
Tested Kilo Code with Qwen3-coder for about 3 hours: CLI showed 23.3K tokens, but Bailian later billed 224.1K across 10 API requests. Suspected Qwen3-coder does not support prompt caching, billing full context per request. Hypothesis only, and it suggests the problem is not exclusive to qwen-code.
Source: https://github.com/QwenLM/qwen-code/issues/83#issuecomment-3112278856 (opens external site)
Did this work for you?What @foxgoblin tried on GitHub (Outcome unclear)
Tested Qwen Code, consumed the free 1M-token quota, then received a debt notice of about 21 RMB and stopped using it. No goal or fix is stated, only the resulting charges.
Source: https://github.com/QwenLM/qwen-code/issues/83#issuecomment-3111815296 (opens external site)
Did this work for you?
Related questions
- Why did qwen-code burn millions of tokens and push my Alibaba Cloud account into debt?
- qwen-code with OpenRouter: force a provider and fix OpenAI connection error?
- How do I enable Agentic Coding Protocol (ACP) support for qwen-code in Zed?
- Ollama qwen3-coder fails with "does not support tools" in qwen-code agent mode
- Does qwen-code support Anthropic-style sub-agents?