AgentPersonaBench
Harbor: pendingA benchmark of 2,460 tasks across 867 traits evaluating latent persona adherence via observable actions in survey, chat, web, and app environments.
- Samples
- 2,460
- Released
- 2026
- Scores
- 7
HSS
Humanity's Sixth Sense
A benchmark for intuitive visual reasoning in MLLMs using diverse image and video inputs with human-written prompts and atomic rubrics.
How to run
ltb · hss
$ npm i -g lettokenburn-cli$ ltb bench info hss
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
13 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 53.6 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| GPT-6.1 Sol | 46.6 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Claude Opus 5.5 | 44.6 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Gemini 3.8 Flash | 41.6 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Claude Fable 5.1 | 40.8 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Gemini 3.7 Flash | 39.8 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Muse Spark 1.3 | 37.4 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Claude Fable 5 | 34.5 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Qwen3.8 Max | 33.0 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Gemini-3.5-flash | 32.6 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Gemini 3.6 Flash | 31.9 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| GPT-6 Sol | 31.2 | N/A | Independent | medium | Oct 6, 2026 | arxiv.org |
| Qwen3.8 Flash | 30.9 | N/A | Independent | medium | Oct 6, 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
HSS evaluates implicit visual inference across temporal, spatial, social, and abstract domains using open-ended tasks. Items are human-authored and verified via multi-stage review to ensure consensus.
Scoring uses atomic rubrics where all criteria must be met for a correct answer. Humans achieve 93.1% accuracy, while the best model (GPT-6-astra) reaches only 53.6%, indicating a significant capability gap.