AGENT MEMORY LEADERBOARD

Leading the AML
agent memory leaderboard

MemoraX ranked #1 among commercial products in both the Textual Memory and Coding Memory tracks of the Agent Memory Leaderboard (AML). The Textual Memory track evaluates factual recall and relational reasoning, while the Coding Memory track measures experience reuse in software engineering tasks

Textual memoryNo.1

MemoraX58.02
System A45.89

12.13 points ahead of the runner-up

Coding memoryNo.1

Task resolution rate

62.00%

Text Memory & Long-Conversation Memory

Recall facts, connect relationships and track information as conversations unfold

LoCoMo-Refined No.1

MemoraX scores 82.65% under the refined judge, 19.05 percentage points above the next system in the repository table. The comparison below uses only that same table

Published score82.65%
Points above the runner-up+19.05%
View source
Published score comparison (%)
  1. MemoraX AI
    82.65%
  2. System A
    63.60%
  3. System B
    58.68%
  4. System C
    58.25%
  5. System D
    48.91%

ScriptMem No.1

MemoraX scores 60.3% with a unified GPT-4o-mini backbone, 17.4 percentage points above the next system in the published comparison

Published score60.3%
Points above the runner-up+17.4%
View source
Published score comparison (%)
  1. MemoraX
    60.3%
  2. System C
    42.9%
  3. System D
    42.0%
  4. System A
    36.4%
  5. System E
    32.6%

BEAM 1M

Evaluating long-term memory, information updates and reasoning across a million tokens of conversation history

Read the benchmark paper
54.08
MemoraX evaluation score

Code Memory & Engineering Experience Memory

Ranked No.1 on SWE-ContextBench, testing cross-task learning and experience reuse in coding agents

SWE-ContextBench

MemoraX achieved a 45% task resolution rate in its evaluation using SWE-ContextBench, which measures how coding agents learn from and reuse experience across related software tasks

Read the benchmark paper
1
Ranked firstSWE-ContextBench
Task resolution rate
45%

Benchmark scope (SWE-ContextBench paper v3)

Base tasks
1,100
Related tasks
376
Repositories
51
Languages
9

Multimodal & Mobile Agent Memory

From mobile interactions to visual-text conversations, MemoraX ranks first across three multimodal memory evaluations

MobileMem · Omni

No.1
34.79pts

Remember everyday interactions. Recall what matters

MobileMem · Struct

No.1
54.72pts

Connect information across apps. Keep up with the user

Mem-Gallery

No.1
76.33%

Connect visual and textual clues across sessions

Explore the results