- Who it helps
- People evaluating scientific reasoning and tool-use strategies.
- Availability
- The public site describes thirty tasks across five worlds and an eight-call budget. Some version-two results are incomplete; version-one reports cannot automatically stand in for them.
Inspect the experiment log
Task records, observations and scoring artifacts help distinguish an informative intervention from repeated probing. A final correct answer may conceal a weak experimental strategy; the separate method score makes that distinction inspectable.
Keep the comparison fixed
Compare agents on the same task version, call budget and controls. Public tasks may have appeared in training data. A deliberately chosen baseline also answers a different question from random probing.
What you can do
- Inspect a task’s log and scoring rules before using its leaderboard as evidence about scientific reasoning. Open reference ↗
Where the evidence stops
No task was executed and no reported score independently reproduced. Incomplete model cells and version differences limit comparisons.
Sources read. Public primary evidence checked October 25, 2026 UTC. No third-party software, model inference or complete product workflow was executed.
Sources & further reading
Web · official
PetriBench task and scoring report ↗
Read tasks, call budget and separate correctness/method scores. Incomplete v2 results remain visible.
Petri Labs · page checked 2026-10-25
Coverage note: Primary material was inspected. Independent execution and matching original YouTube, X and Reddit discussion remain incomplete; unavailable engagement counts are omitted.
The record over time (1)
Dated source checks and published explanations. An observation date can differ from the original release date.
- · Added to Current
PetriBench evaluates agents on small scientific worlds with a limited number of experimental calls. It records correctness separately from how the agent chose and used experiments.