AGI Index dimension · weight 6%
Memory
Retention and retrieval across long interactions.
Whether a system holds relevant state over extended interaction and retrieves it accurately when it matters — as opposed to holding a long context window and attending to the wrong part of it.
The argument
Evidence and counter-evidence
Both sides are published at equal weight. A framework that only records what raises a score is not measuring anything.
Raises the score
- Effective context has grown by orders of magnitude and retrieval within it has become far more reliable.
- Persistent memory across sessions is now a shipped product feature rather than a research demonstration.
Holds it down
- Retrieval accuracy degrades in the middle of long contexts in ways users do not detect.
- Systems are poor at forgetting — distinguishing what should persist from what should not is largely unsolved, and it matters more as windows grow.
What the score reads from
Long-context retrieval
Cross-session consistency
Related coverage
Everything bearing on memory
News
Research
Argument
Context
Where this sits against the rest
- Multimodality82
- Coding81
- Tool use76
- Reasoning74
- Generalization66
- Planning61
- Learning59
- Autonomy53
- Social intelligence52
- Scientific discovery48
- Embodiment41
How these scores are producedMethodology v1.2