CritPt
Harbor: pendingCritPt — AI evaluation benchmark listed by opencompass.
Verified71%Knowledge & Reasoning
- Samples
- 70
- Released
- 2025
- Scores
- 141
ProPerMem
ProPerMem benchmark for memory-driven proactive assistance
Evaluates memory-driven proactive assistance using egocentric video and ~618K token year-long memories, scoring content and timing of interventions.
How to run
ltb · propermem
$ npm i -g lettokenburn-cli$ ltb bench info propermem
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
6 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Claude Sonnet 5 | 15.0 | rag | Independent | medium | Oct 15, 2026 | arxiv.org |
| GPT-5.3 | 10.0 | rag | Independent | medium | Oct 15, 2026 | arxiv.org |
| Gemini 3.1 Flash Lite | 8.5 | rag | Independent | medium | Oct 15, 2026 | arxiv.org |
| Qwen3.8 27B | 7.0 | rag | Independent | medium | Oct 15, 2026 | arxiv.org |
| VideoLLM-online | 1.6 | rag | Independent | medium | Oct 15, 2026 | arxiv.org |
| MMDuet2 | 0.0 | rag | Independent | medium | Oct 15, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsLoading community runs…
About
Pairs egocentric video with distractor-rich, year-long user memories (~618K tokens) and human-verified trigger windows. Includes clips requiring silence.
Scores strict accuracy: one grounded response inside the window. Baselines perform poorly (<15%), while oracle evidence yields 96%, highlighting timing and retrieval challenges.