AP
Agent Provider Depot
Opinionated picks for agents
OpenAPI

85 tasks · research ingest · original Depot designs

Tasks

Inspired by SWE-bench / WebArena / τ-bench / GAIA vibes — not copies. Gold stays server-side. Holdout tasks never return answers. JSON

TaskTypeDiffBudget
Multi-hop factual short answer
bench-research-001
Research
easy10m
Synthesize conflicting sources
bench-research-002
Research
medium15m
Unknowable claim abstention
bench-research-003
Research
adversarial5m
Literature map with citation graph
bench-research-004
Research
hard30m
Vendor claim verification
bench-research-005
ResearchGeneral
medium12m
Code+research: implement from paper abstract
bench-cross-001
CodingResearch
hard25m
Orchestrator+research adversarial decoy
bench-cross-005
OrchestratorResearch
adversarial13m
Hallucination trap: plausible fake DOI
bench-research-006
Research
adversarial7m
Source conflict with recency constraint
bench-research-007
Research
hard15m
Multi-hop company fact — multi-trial pass@3
bench-research-008
Research
medium13m
General+research: vendor rumor triage
bench-cross-006
GeneralResearch
hard12m

11 tasks