Our summary
A high test score says little about price or ease of access. This comparison gives quality forty points, then adds probability calibration, access, price, answer types and development tools. The retrieved page covers thirteen hosted models, up from nine in an earlier version.
For quality, the site scales independent results from three test boards and averages the boards that measured each model. Its stated method excludes vendor-run results. Models with no independent score remain unranked, while models you must host yourself appear separately because access and costs differ.
Those weights are the site's choices, not a universal performance order. One board gives less coverage than three, and displayed ranks do not consistently match the totals in this snapshot. Before choosing, read the underlying results and current access terms rather than relying on the table's position alone.
Key takeaways
- Find out whether one board or several measured a model. The combined score hides that difference in coverage.
- Read price, access and tools separately from test performance, even though all contribute to the site's total.
- Compare the totals yourself: this snapshot's table order does not consistently follow its displayed numbers.