GBrain Memory System Achieves 8.3x Win Rate in Benchmark Tests

robot
Abstract generation in progress

AIMPACT News, on May 7 (UTC+8), a benchmark test released on May 7, 2026, shows that the open-source graph memory system GBrain outperforms OpenClaw’s QMD setup, achieving an 8.3x win rate with a record of 58 wins and 7 losses on 352 files of the AI Heroes knowledge base slices and 150 question evaluation sets.
The evaluation set includes 65 difficult, 53 medium, 32 easy, 30 cross-source, and 25 discriminative questions.
GBrain demonstrates better performance in cross-source question recall, latency, and handling complex connections through hybrid retrieval (BM25 + vector embeddings) and a typed link graph extractor.
The preset rule judgment is “partially passed,” as the average P@5 difference (+0.081) is below the migration threshold of +0.15, but the pattern is clear, and a comprehensive migration plan has been scheduled.
(Source: InFoQ)

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin