About: a public review database and a cold-start experiment

The problem

Public information on Chinese multimodal models is fragmented: launches claim “first”, leaderboards give one number, and technical reports use incompatible protocols. Buyers need to know who is steadier in their scenario, at what price, with what risk.

We break 32 models into eight comparable dimensions, keep the evidence and confidence behind each, and place price, context and licence in one table. You do not have to trust our ranking — you can re-weight it for your own business.

Independence

We are not affiliated with, sponsored by, or paid by any vendor. We do not sell score improvements or run affiliate promotions. Any future commercial placement will be labelled and fully separated from scoring.

Scores are set by the editorial desk against a fixed rubric. AI assists with research, translation and drafts; final scoring, fact-checking and wording remain human.

Why an English edition

To give global developers a realistic view of Chinese models beyond leaderboard copying — and because bilingual pages are part of the experiment: how hreflang, duplication and crawl budget behave on a new domain.

About mistakes

We will get things wrong: a version, an expired price, a vendor claim mistaken for third-party evidence. Our rule is verifiable, traceable, never silently edited. Email jizhi2024@foxmail.com with evidence; we reply within seven days and log the correction.