Volcano Engine Coding Plan exposed to double billing: billed supposedly by the number of times, but actually also deducted based on token consumption converted into counts

According to monitoring by 1M AI News, users have discovered that the billing method for Volcanic Engine’s Ark Coding Plan involves hidden mechanisms that are not fully disclosed. A developer on V2EX reported that the quota consumption rate is much faster than that of similar plans on other platforms; after contacting customer service, they received a reply: “If the tokens consumed per single model call are significantly higher than the average consumption, the estimated number of calls that can be made within the cycle will also be far less than 6,000.”

Taking the Pro plan (200 yuan/month) as an example, the nominal quota is 6,000 requests every 5 hours, 45,000 requests per week, and 90,000 requests per month. However, the actual billing does not deduct one token per call; instead, it is converted into multiple deductions based on the token consumption of each individual call. The conversion formula used by users is: usage = max(round(use_token/token_limit), 1). Different models have different hidden multipliers: DeepSeek-V3.2 about 2x, Doubao-Seed-2.0-Code about 4x, Doubao-Seed-2.0-Pro about 6x. In other words, a single call using Doubao-Seed-2.0-Pro may be counted as six quota deductions.

The user provided an example: one call consumed 510,000 tokens; on platforms like Alibaba Bailian and others, it would only count as one, but on Volcanic Engine it might be converted to about 20. When AI programming agents perform complex tasks, consuming hundreds of thousands or even tens of thousands of tokens per call is common. This billing method can quickly deplete the plan’s quota.

Currently, domestic Coding Plan plans generally charge based on the number of calls. Platforms such as Alibaba Bailian, Xiaomi MiMo, and others deduct once per call and do not convert based on tokens. The “count-based, token-converted” dual-layer billing method used by Volcanic Engine is relatively rare in the industry, and it is not prominently disclosed on the plan page. Users only learn about this mechanism after experiencing abnormal consumption and contacting customer service. ByteDance’s AI programming tool Trae has also recently been reported by users to have shifted from pure call-based billing to a similar token-based conversion method for counting calls.