Method Drift›Agent / long-term memory
MAGMA
MAGMA: A Multi-Graph based Agentic Memory Architecture for AI AgentsAgent / long-term memory · first seen Apr 16, 2026
superseded — cited as a baseline and beaten by newer methods
1 papers critique it · 1 beat it on benchmarks
What papers say
Verbatim critique sentences, each from a paper that cites MAGMA as a baseline.
“Furthermore, even when continuous scores or edge weights are introduced, retrieval is still largely governed by fixed similarity search, manually designed scoring functions, or static heuristic traversal rules.”
— HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
Beaten on benchmarks
Head-to-head results where a newer method reports beating MAGMA. Values are copied from the source paper's tables — verify against the cited paper.
- HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE beats MAGMA · Overall [gpt-4o-mini]
0.739 vs 0.700
- HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE beats MAGMA · Overall [Qwen2.5-3B]
0.548 vs 0.499
- HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE beats MAGMA · LLM Score [GPT-4o-mini]
0.824 vs 0.807
- HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE beats MAGMA · F1 [GPT-4o-mini]
0.678 vs 0.640
- HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE beats MAGMA · LLM Score [Qwen2.5-3B]
0.527 vs 0.424
- HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE beats MAGMA · F1 [Qwen2.5-3B]
0.429 vs 0.337
- HAGE: Harnessing Agentic Memory via RL-Driven Weighted Graph Evolution
HAGE beats MAGMA · Avg. Score [all]
0.739 vs 0.700
Newer alternatives
Recent methods in the same sub-problem, not yet superseded in the knowledge base.
- May 11, 2026
- Mar 16, 2026
- Nov 21, 2025