AI Index
Benchmarks
The evaluation layer, and how much of it is self-reported. 6 entries.
Sort:
BenchmarkBEAMExtreme-scale conversational memory benchmark (1M-10M tokens); tests whether memory matters at frontier context windowsBenchmarkDolphinBenchMem0's own benchmark; provenance=self_reported per index policyBenchmarkLongMemEval v2Extension of LongMemEval toward 'experienced colleagues' recall+reasoning over richer agent tracesBenchmarkMarathonMemBenchLong-horizon memory benchmark for LLM agents; positions itself against LongMemEval as the 'canon' multi-session suiteBenchmarkMemoryAgentBenchICLR 2026; incremental multi-turn evaluation; the only one with a reproducible public harness
★ 486 on GitHub
BenchmarkSTATE-BenchTests task completion, not just fact retrieval; complement to retrieval-only benchmarks