Still running
Fourteen of these competitions are open. Public ranks move, and the private board is the only one that counts. Everything below is a snapshot with a date on it.
Kaggle is the setting we keep coming back to, because it is one of the few places an agent loop can be wrong in public: thousands of humans on the same problem, a deadline that does not negotiate, and a private leaderboard that arrives after every decision has already been made.
What was run
Built on Humanize 2, we designed roughly ten distinct agent workflows and pointed all of them at live competitions. They differ in the things we think actually matter on a long run:
- how context is managed across hours of work,
- how several skills are orchestrated inside one run,
- how much of an end-to-end machine-learning engineering pipeline the agents own,
- whether a second agent reviews the first one's work, and
- whether attempts run in parallel or in sequence.
Where they landed
Completed — nineteen competitions. Fourteen finished in the top 5%, including five results at or inside the top 1%.
Ongoing — fourteen official competitions. Six currently sit in the top 5%, four of them in the top 3%, and the numbers are still moving.
For comparison, Codex on GPT-5.5 at xhigh reasoning effort generally peaked at around the top-5% level across most of the same competitions. Same models, different arrangement, and the arrangement is worth the tail of the distribution.
What we think is doing the work
Our preliminary read is that the gains come primarily from the diversity of reasoning strategies that multi-agent collaboration makes available. A single strong agent narrows early and commits; several agents arranged to disagree keep more than one line of attack alive long enough to be tested against a validation split that has not been quietly bent to fit.
That is a hypothesis with a number attached, not a conclusion. Further analysis is ongoing.
What these figures are, and are not
A finish and an estimate are two different claims, and we do not report them as one. A finish is an exact position on a closed leaderboard. An estimate is a score, submitted after the competition closed and placed against the frozen final board — real as a score, and not a rank, a medal, or evidence of having competed. Both appear above; the audit says which is which, one row at a time.