The AI Parity Index is a public record of how AI performs against measured human baselines, task by task, with receipts.
Most public claims about AI and work rest on predictions or usage logs. The Index tests instead: AI takes the real tasks of the economy, is scored against a measured human baseline, and every run is published so anyone can re-check it. The result is a lower bound. Everything on the board is proven, nothing is projected.
| Recursiv Labs | built this prototype, owns the method as it currently stands, and runs the tests. Recursiv is the measurement instrument, not a contestant, and is never ranked. |
| SPARK AI Research | the proposed publisher, not the current one. An index like this needs a neutral academic home, and SPARK at the San Diego Supercomputer Center is the one we would want. Nothing here has been reviewed or adopted by SPARK or UC San Diego yet; this draft exists to ask them whether it should be. |
| Sponsors | would fund verification for an industry, be named on the page they funded, and have no say over scores. None exist today. |
| Scores | no funder, sponsor, or platform can change a score. The pass criterion is stated before a run and executed, not judged. |
| Data | verification runs use public materials only. Receipts are published with every number. Private data never enters the Index. |
| Revisions | past readings are never revised. Method changes carry a version number. |
| Measure your work | Have your own operation measured with the same method, inside your walls. Recursiv Labs runs the engagement. Work with us › |
| Get on the board | If you ship a model or an AI product, ask for it to be tested. Open a request › |
| Sponsor a vertical | Fund the tests for an industry. Your name goes on the page; you get no say over the scores. Write to us › |
The Index sits alongside the measurement community and cites it: the Anthropic Economic Index tracks usage, the Stanford AI Index surveys the year, the OECD estimates exposure, and MIT’s AI Labor Index scores expert predictions. Those projects watch and predict. This one tests.
The method, the numbers, and every mistake in them belong to Recursiv Labs. Questions about either go there.
Working prototype by Recursiv Labs · not an official SPARK AI or UC San Diego publication · Data · © 2026