ExplorationBench places agents in two executable worlds whose rules conflict with familiar knowledge. AlienCode and AlienLogic provide flawed manuals, environment feedback and tool schemas, then score held-out tasks with an interpreter or proof checker. Across 10 systems, the strongest trajectories learned and applied unfamiliar rules, but outcomes varied widely and additional exploration sometimes erased earlier gains. The benchmark covers 55 discovery targets and 140 tasks; its site states that code and task data are not being released.
ExplorationBench tests whether agents can learn unfamiliar rules
2026-09-25Frontier watchWATCH
ExplorationBench places agents in two executable worlds whose rules conflict with familiar knowledge. AlienCode and AlienLogic provide flawed manuals, environment feedback and tool schemas, then score held-out tasks with an interpreter or proof checker. Across 10 systems, the strongest trajectories learned and applied unfamiliar rules, but outcomes varied widely and additional exploration sometimes erased earlier gains. The benchmark covers 55 discovery targets and 140 tasks; its site states that code and task data are not being released.
Availability: project site says code and tasks will not be released
Imported from the existing research archive. Migration preserves the recorded judgment and is not a fresh verification.
What this means for you
For research-agent teams, this separates recalling known science from learning through experiments more cleanly than ordinary knowledge benchmarks. Watch it until tasks and code are available; without them, independent reproduction and contamination checks remain limited.
FRONTIER WATCH
Public artifacts and validated results are different.
The recorded fact is the public work and its artifacts. Author-reported outcomes do not establish independent reproduction. Next checks include usable code, task conditions and third-party validation.
Evidence & related material
Technical details & scores
| Worlds | AlienCode and AlienLogic |
|---|---|
| Scope | 55 discovery targets; 140 held-out tasks; 10 systems |
| Verification | executable interpreter, proof checker and bounded certifier |
| Best Trajectory | AlienCode rose from at most 15.7% before exploration to 87.6% after four rounds |
| Failure Mode | continued exploration can stall or reverse gains |
| Availability | project site says code and tasks will not be released |
Technical impact 3
Engineering value 3
Adoption signal 1
Maturity 1
Verification cost 4
Risk 2