AgentsHow we source and rate
We cite only three kinds of sources: official vendor docs and pricing, independent benchmarks, and reputable press.
Rules
- Every price and feature links to an official page, with the date we checked it.
- Vendors can't pay for placement or rankings.
- We don't rely on social posts, anonymous reviews or affiliate listicles.
- Benchmark numbers belong to their publishers. We link to them rather than republishing them.
Benchmarks we trust
- SWE-bench Verified: Can a model fix real GitHub issues? The standard for coding agents.
- OSWorld: Can an agent complete open-ended tasks on a real computer?
- GAIA: General AI-assistant tasks needing reasoning, browsing and tools.
- HAL (Princeton): Reproduced agent evaluations, including GAIA.
- Artificial Analysis: Independent model intelligence, speed and price leaderboards.
Directory data
Directory recipes come from the MIT-licensed botdirectory.ai feed, with attribution on every page.