Our summary
A model's test score does not tell you whether it is affordable or easy to access. This comparison gives quality forty points, then adds probability calibration, access, price, answer types and development tools. The retrieved page lists thirteen hosted models, rather than the nine in an earlier version.
For quality, it scales independent results from three different test boards and averages the boards that measured each model. Vendor-run results are excluded under the stated method. Models without independent scores stay unranked; models you must host yourself are listed separately because their costs and access differ.
The total reflects this site's chosen weights, not a universal performance order. A score based on one board has less coverage than one based on three. The displayed rank order also does not consistently follow the displayed totals. Read the underlying results and current access terms before choosing a model.
Key takeaways
- Check whether a model was measured by one board or several; the combined score alone hides that difference.
- The site's total includes price, access and tools as well as test performance. Read those parts separately.
- Compare the displayed totals yourself. In this snapshot, the table's order does not consistently follow those numbers.