How we verify
Every item points to its original source. We do not invent facts, headline numbers, or authors, and a listing is not an endorsement. If we cannot confirm…
Our promise is simple
Every item points to its original source. We do not invent facts, headline numbers, or authors, and a listing is not an endorsement. If we cannot confirm something, we leave it blank rather than guess.
Where model details come from
Company and model facts — who makes it, when it shipped, its licence, price and context window — come from the model maker's own model card, documentation, or announcement. We link that source on the comparison so you can read it yourself.
Where benchmark scores come from
Scores are copied from the benchmark's own published results or from the model maker's report. We do not run the benchmarks ourselves. Vendors report their own numbers, testing conditions differ, and a missing score is left blank — it is not a zero and it does not mean the model failed.
Why numbers move
Benchmarks are revised, models are updated, and evaluation settings change. Treat every score as a dated snapshot and check the original before making a decision. The benchmark to trust is the one that measures your task, tested with your own examples.
What we do not do
No paid placements. No edited scores. No ranking that claims one model is best for everyone. We show what was published and where it came from, and we let you compare.
How to correct us
If something here is wrong, use Report on the item and tell us what is incorrect. Reports go to a small private review queue — not to the public. We fix the record or remove it, and we would rather hear from you than publish something misleading.