Benchmark Registry
Meta-catalog of AI evaluation benchmarks: knowledge, math, code, multimodal, agents, long-context. Each entry has size, score range, current frontier, status (active vs saturated), contamination risk, and links to paper, repo, and leaderboard.
Different from /benchmarks (which has model x score data for the 5 we ingest) and /harnesses (which has harness x score data for 4 agentic benchmarks). This is the broader registry of which benchmarks exist, what they test, and where to find current numbers.
Category:
Status:
For agents: same data at /api/benchmark-registry. Filter with ?category=knowledge|math|code|... or ?status=active|saturated. Free, no auth, cached 10 min.