Spaces:

---

gaia-benchmark

/

leaderboard

Like 628

Running on CPU Upgrade

Hi GAIA maintainers,

Thanks for maintaining the GAIA benchmark and leaderboard.

I noticed that the current leaderboard appears to contain many duplicate or near-duplicate submissions. In several cases, the same method, team, or model seems to appear multiple times with very similar configurations and scores. This makes the leaderboard harder to read and makes it more difficult to identify the actual state of the art.

Would it be possible to add a deduplication or filtering policy, for example:

Show only the best submission per method/team/model by default.
Keep all historical submissions accessible in an expanded or “all submissions” view.
Group submissions by submitter, method name, or model family.
Mark entries that are superseded by a later submission from the same team/method.

I can share concrete examples of duplicate-looking entries if helpful.

Thanks again for maintaining the benchmark.