List your company

Single-Hop Fact-based Reasoning on MAB FC-SH 262K v3 (test)

93AccuracyHeadline pipeline
3.5626.785073.22May 31, 2026

Evaluation results

MethodLinks
Headline pipelineBackbone=gpt-4o2026.0593
Headline pipelineBackbone=gpt-4o-mini2026.0582
Ablation ABackbone=gpt-4o-mini, Pipeline architecture=chunk-40962026.0573
LLM-judgment baselineBackbone=gpt-4o-mini2026.0561
GPT-4o (long-context)Backbone=gpt-4o, Pipeline architecture=long-context2026.0560
HippoRAG-v2 (best published)Backbone=gpt-4o-mini2026.0554
BM25Backbone=gpt-4o-mini, Pipeline architecture=MAB chunk-512 default2026.0548
Zep / GraphitiBackbone=gpt-4o-mini, Pipeline architecture=temporal KG2026.057