1. What gets included
A model must be built by an Asian company, institution or research lab and have enough public first-party information to describe its purpose and access. AIOAI currently favors models that give users a meaningful regional or task-specific alternative.
2. What the scores mean
Overall score is a broad discovery score across capability, practical access and regional relevance. Value score considers how much useful access and model breadth a user can obtain. Task scores answer a narrower question: how suitable does the current evidence make this model for coding, writing, image, video, voice or translation work?
3. What the scores do not mean
They are not controlled benchmarks, audited price comparisons, user-review averages or guarantees of availability. A score should help you form a shortlist, not replace testing on your own prompts and market.
How the current order is calculated
The stored overall, task and value numbers are provisional editorial estimates inherited from the initial index, not measurements. Best overall uses the task estimate when a task is selected, otherwise the overall estimate. Best value uses the value estimate alone without a task; with a task it rounds 65% of the task estimate + 35% of the value estimate. Open weights filters eligibility and keeps the same task or overall order. Equal scores keep the dataset order.
Publishing this calculation does not validate the input estimates. We have not published a repeatable test set or version-specific score evidence. In particular, 99/100 does not establish that a model is the cheapest or objectively better. Do not compare a video product with a coding model on this scale.
Official-page checks are not hands-on tests
Selected profiles now distinguish the release mentioned in official documentation, the web product and its API route. Their dated source notes are public-page checks, not account creation, paid generation or payment verification. A newer version note does not mean the legacy score has been retested. Unverified quota, watermark, registration and payment conditions remain labelled as unverified.
4. Source and review policy
Profiles link to official product pages, documentation or research pages. AIOAI records a dataset review date and avoids unsupported labels such as “trending” when no repeatable measurement exists. Pricing, free plans, regional registration and API access can change, so the official source remains authoritative.
5. Known limitations
Coverage still differs by region and is not a catalogue of every checkpoint or minor model version. Some entries are consumer tools; others are research releases or model families. Direct score comparisons should therefore be treated as orientation, especially across different product types.