Fragmented Pricing Models for Domestic AI Models: Only a Few Charge by Token

On 2026-11-03 at 18:05, the AI section of DEV Community published a piece titled that domestic Chinese AI models have messy pricing schemes, with only a few charging by token. The author started from a simple goal: building a table that answers which Chinese AI model is cheapest for a given task. About a day in, he concluded that the question itself does not hold up. Among the 111 models he normalized, only 15 charge per token. The rest charge per image, per generation, or a flat rate per call. The article presents the distribution from the feed it rebuilds daily: 49 models charge a flat rate per call, roughly 44% of the total; 33 charge per generation, about 30%; and 15 charge per 1M tokens. On that basis, the author argues that comparing vendors by their headline price numbers amounts to comparing nothing, and that this split in billing units creates structural obstacles for cross-platform price comparison and for integrating these models into tools and workflows. The piece does not name the individual models behind each category, and it does not rank one pricing scheme as better or worse than another. As of the publication, the picture rests on the author's own normalization effort and his daily-rebuilt feed, with no follow-up from the vendors included.