©2026 adebench · adecubedbuilt 2026-11-04 from the reports in the repo

← leaderboard

supermemory

memory and context engine: an LLM agent extracts and versions memory entries from documents. https://github.com/supermemoryai/supermemory

Core
74% · 40.5 / 55 (mean)
Core (public2)
public2 run in progress
Private set
after the public2 runs
Full
78% · 50.5 / 65 · 65 of 100 measured
Runs
5 · core 38.1–43.0 · the details below are the median run
Common with gbrain
51.3 vs 50.4 on 65
Chars per answer
405 · answer after 54
Write → visible
2.1 s · p95 2.1 s
Write-back
0 of 2 degraded answers came back
Reached through
REST
Model on write
yes: an LLM agent extracts every document
Key needed
LLM provider
Where it runs
self-hosted
License
MIT
Deterministic
no: extraction is a model call
Configuration
Versionsupermemory-server 0.0.8 (windows-x64 release binary)
Modeself-hosted one-binary server (supermemory local), REST, container 'adebench'
DoorPOST /v4/search {q, containerTag: "adebench", limit: 10, include: {documents: true}}, searchMode memories (the default)
Resultslimit 10 and threshold 0.6 (API defaults, threshold not sent), no cut
Model insidegemini-3.1-flash-lite-preview (native Gemini provider, fixed by the binary): a memory agent extracts memory entries from every document and decides which entry updates which
EmbeddingsXenova/bge-base-en-v1.5, local, 768 dimensions (the self-hosted default)
Storagea scratch server data directory (format not stated)
Not defaultnone (the import sends one document per call: a batch of 21 stalled on Windows with 0.0.8)

Notes: self-hosted server 0.0.8, native Gemini provider (gemini-3.1-flash-lite-preview, fixed by the binary), local bge-base-en-v1.5. Run 2026-09-30, door memories, fingerprint e37cca2b5e. Report: examples/supermemory_report/reference.md · All runs: examples/supermemory_report/repeats.json

Sections
door
18.8
of 25 · 6 pass · 1 fail
FAIL Which version of MailBridge is installed for the mailbox? — STALE value delivered next to the current one: 1.3.0
ERROR What is the owner's phone number? — empty answer from the door
cards
not measuredsection not measured: no entity cards in this memory
updates
10.0
of 10 · 3 pass · 0 fail
time
5.0
of 10 · 3 pass · 3 fail
FAIL semantic memories carry their age (36/42) — 86%
FAIL an imported memory reaches the door with its original date — imported memory not served within 30 s
FAIL «what did pc2 do?» finds the episodes signed [pc2] — 0 episodes in the answer
live state
10.0
of 10 · 7 pass · 0 fail
abstention
7.5
of 10 · 2 pass · 2 fail
FAIL Which port does the Fulmicotone service use? — 1 keyword facts for something that does not exist
FAIL What does the calendar plugin Girandola do? — episodes presented as direct matches; 6 keyword facts for something that does not exist
file search
not measuredsection not measured: no repo configured (--repo)
graph
not measuredsection not measured: no case with evidence (all SKIP or no cases)