RHELM: Benchmark for Realistic and Evolving Long-Horizon Memory

RHELM is a benchmark dataset for evaluating long-horizon memory capabilities in AI assistants. It was created by Microsoft and last updated on June 1, 2026. The benchmark provides realistic, heterogeneous, and temporally evolving memory sources.

Use Cases

• Benchmarking multi-hop reasoning based on heterogeneous memory sources.
• Evaluating temporal synthesis capabilities based on evolving memory narratives.
• Testing hallucination detection in AI assistants based on realistic, long-horizon dialogues.

Strengths

• Designed to address limitations of static dialogue benchmarks.
• Focuses on realistic, heterogeneous, and temporally evolving memory sources.
• Created by a major research organization (Microsoft).

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update timestamp is in the future (2026-06-01).

Provenance

Source

Microsoft

Collection Method

Likely constructed for benchmarking purposes, but specific collection method is unknown.

Freshness

Last updated 2026-06-01 08:31:55