I ran tests against their Qwen LLM API (code completion + chat) on a project that’s ~3,200 lines of code. Over a period of 2 hours, their system claimed I had consumed 10 million tokens.
To put this into perspective:
Even if the entire 3,200 lines were repeatedly sent as context (≈150k tokens max per call), it would take ~67 calls to reach 10M tokens.
My logs show nowhere near that volume of requests.
Yet their billing immediately deducted ~$4.19 and left me with a “pending” $7.18 outstanding charge.
When I raised this with support, their response was:
“This may be because of plugins like VSCode auto-completing with long contexts” or “Every call carries historical context, so tokens add up faster.”
That explanation doesn’t pass even basic sanity checks:
Carrying context doesn’t magically multiply usage by 100x.
No LLM API I’ve used (OpenAI, Anthropic, Mistral, Groq) has ever produced token consumption so detached from real usage.
Suggesting that VSCode plugins are responsible is a cop-out, not a root cause analysis.
Instead of accountability, they tried to silence me with $200 in coupons valid only for Alibaba Cloud. I refused. I’ve since demanded:
Refund of the $4.19 they already charged.
Clearing of the fraudulent $7.18 “outstanding” bill.
Deletion of my account and confirmation my code/logs are erased (privacy concerns).
They refuse to provide (2) and (3).
This raises two much bigger issues:
Trust: If their billing is this opaque, how can any developer safely use their API at scale?
Privacy: If they’re happy to gaslight customers on token usage, what are they doing with uploaded code and prompts behind the scenes?
I’ve started documenting and publishing evidence across platforms (Reddit, LinkedIn). Next step: reaching out to journalists.
Has anyone else faced similar billing or support failures with Alibaba Cloud? Would love to hear from engineers who’ve tested their API billing in detail.
—
Warning to devs: be extremely cautious using Alibaba Cloud for AI workloads. Opaque billing + impotent support = dangerous combination.