Start from your modality

Memory problems look different for a robot, a video assistant, or a phone agent. Pick the setting you work in to see how its systems remember and how they are evaluated.

Image
64 systems · 19 benchmarks
T1 18 · T2 39 · T3 7Most common form: Structured records & graphs
Video & streaming
84 systems · 23 benchmarks
T1 27 · T2 54 · T3 3Most common form: Retained source evidence
Audio & speech
8 systems · 10 benchmarks
T1 1 · T2 7 · T3 0Most common form: Latent state
GUI & computer use
41 systems · 7 benchmarks
T1 13 · T2 25 · T3 3Most common form: Structured records & graphs
Embodied & 3D
100 systems · 23 benchmarks
T1 17 · T2 79 · T3 4Most common form: Structured records & graphs

Management tiers by modality

Management tier by modality, shown as shares so modalities of different sizes compare fairly. Switch to Stacked for counts.

Image 100%
Video & streaming 100%
Audio & speech 100%
GUI & computer use 100%
Embodied & 3D 100%

Systems by modality and management tier
Category	T1 Static Memory	T2 Dynamic Memory	T3 Self-Evolving Memory
Image	18	39	7
Video & streaming	27	54	3
Audio & speech	1	7	0
GUI & computer use	13	25	3
Embodied & 3D	17	79	4

Where memory is stored

Storage and indexing structures by modality. A system with several stores counts once in each.

Flat sequential records 106
Vector-indexed collection 94
Graph / relational store 102
Hierarchical multi-tier store 107
Modular adapter bank 27

Systems by storage structure and modality
Category	Image	Video & streaming	Audio & speech	GUI & computer use	Embodied & 3D
Flat sequential records	16	36	4	13	37
Vector-indexed collection	33	20	2	13	26
Graph / relational store	20	16	3	15	48
Hierarchical multi-tier store	24	35	3	20	25
Modular adapter bank	5	7	0	2	13

Embodied and video lead; audio is just starting

Cumulative systems by primary modality and year of the latest paper version.

Jan 2027 · 297 systems

Image 64

Video & streaming 84

Audio & speech 8

GUI & computer use 41

Embodied & 3D 100

2023202420252026