# Humanity's Sixth Sense: Benchmarking Intuitive Visual Reasoning in Multimodal Models (Humanity's Sixth Sense (HSS))

> HSS tests what a model can infer from images and silent videos across temporal, spatial, social, and abstract reasoning. We show the published results as display-only evidence, excluded from overall and category rankings.

Canonical page: https://benchlm.ai/benchmarks/scale-hss

- Category: [Multimodal & Grounded](/multimodal-grounded)
- Last updated: October 15, 2026 snapshot

## About Humanity's Sixth Sense (HSS)

- Year: 2026
- Tasks: 522 tasks (288 image, 234 video)
- Format: Pass@1 (%) averaged over three attempts per task
- Difficulty: Intuitive visual reasoning
- Paper: [Humanity's Sixth Sense paper](https://labs.scale.com/papers/humanitys-sixth-sense)
- Source: [HSS leaderboard](https://labs.scale.com/leaderboard/hss)
- Source: [HSS paper](https://labs.scale.com/papers/humanitys-sixth-sense)
- Source: [HSS dataset](https://huggingface.co/datasets/ScaleAI/HSS)

The release contains 522 tasks: 288 images and 234 videos, organized into four domains and 11 subdomains. Pass@1 averages three attempts per task. An answer passes only when it meets every rubric criterion. The published ± values are half-widths of 95% bootstrap intervals over tasks.

Compare the published model configurations and their reasoning settings. The human baseline is 93.1%; overlapping uncertainty intervals do not establish a reliable ordering.

Video inputs exclude audio and transcripts. Claude Opus 5 grades the answers, and the scores depend on that judge. Public-web media may have appeared in training data; social judgments measure agreement with annotators. These results are display only and do not affect overall or category rankings.

HSS results are display only and excluded from overall and category rankings. The dataset annotations use the MIT License; the images and videos retain their original rights and separate research-use conditions.

Humanity's Sixth Sense (HSS) is currently displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.

## Leaderboard (27 models)

| Rank | Model | Creator | Score |
|------|-------|---------|-------|
| 1 | [Gemini-4-Argon (high)](https://labs.scale.com/leaderboard/hss) | Google | 55.40% ±4.20 points |
| 2 | [GPT-6-Astra (max)](/models/gpt-6-astra) | OpenAI | 53.60% ±4.10 points |
| 3 | [GPT-6.1-Sol (max)](/models/gpt-6-1-sol) | OpenAI | 46.60% ±3.80 points |
| 4 | [Claude-Opus-5.5 (xhigh)](/models/claude-opus-5-5) | Anthropic | 44.60% ±3.80 points |
| 5 | [Muse Wonder 1.0 (max)](https://labs.scale.com/leaderboard/hss) | Meta | 43.70% ±4.20 points |
| 6 | [Gemini-3.8-Flash (high)](/models/gemini-3-8-flash) | Google | 41.60% ±3.70 points |
| 7 | [Claude-Fable-5.1](/models/claude-fable-5-1) | Anthropic | 40.80% ±3.60 points |
| 8 | [Gemini-3.7-Flash (high)](/models/gemini-3-7-flash) | Google | 39.80% ±3.70 points |
| 9 | [Muse Spark 1.3 (max)](/models/muse-spark-1-3) | Meta | 37.40% ±3.50 points |
| 10 | [Claude-Fable-5 (max)](/models/claude-fable) | Anthropic | 34.50% ±3.70 points |
| 11 | [Qwen3.8-Max (max)](/models/qwen3-8-max) | Alibaba | 33.00% ±3.60 points |
| 12 | [Gemini-3.5-Flash (high)](/models/gemini-3-5-flash) | Google | 32.60% ±3.40 points |
| 13 | [Gemini-3.6-Flash (high)](/models/gemini-3-6-flash) | Google | 31.90% ±3.40 points |
| 14 | [GPT-6-Sol (max)](/models/gpt-6-sol) | OpenAI | 31.20% ±3.40 points |
| 15 | [Qwen3.8-Flash (max)](/models/qwen3-8-flash-next) | Alibaba | 30.90% ±3.30 points |
| 16 | [Claude-Opus-5 (max)](/models/claude-opus-5) | Anthropic | 30.50% ±3.50 points |
| 17 | [GPT-5.6-Sol (max)](/models/gpt-5-6-sol) | OpenAI | 30.00% ±3.30 points |
| 18 | [GLM-5.3-Flash (max)](/models/glm-5-3-flash) | Z.AI | 26.80% ±3.00 points |
| 19 | [Kimi-K3 (max)](/models/kimi-k3) | Moonshot AI | 25.50% ±3.10 points |
| 20 | [MiMo-v2.6-Flash (high)](/models/mimo-v2-6-flash) | Xiaomi | 25.40% ±2.90 points |
| 21 | [MiMo-v2.6-Pro (high)](/models/mimo-v2-6-pro) | Xiaomi | 25.20% ±2.90 points |
| 22 | [Claude-Sonnet-5 (max)](/models/claude-sonnet-5) | Anthropic | 24.80% ±3.20 points |
| 23 | [Muse Spark 1.2 (xhigh)](/models/muse-spark-1-2) | Meta | 24.50% ±3.10 points |
| 24 | [Claude-Opus-4.8 (max)](/models/claude-opus-4-8) | Anthropic | 23.50% ±3.20 points |
| 25 | [GPT-5.6-Terra (max)](/models/gpt-5-6-terra) | OpenAI | 22.70% ±3.20 points |
| 26 | [GPT-5.6-Luna (max)](/models/gpt-5-6-luna) | OpenAI | 21.60% ±3.20 points |
| 27 | [GPT-6-Luna (max)](/models/gpt-6-luna) | OpenAI | 21.00% ±3.00 points |

## FAQ

### What does Humanity's Sixth Sense (HSS) measure?

HSS tests what a model can infer from images and silent videos across temporal, spatial, social, and abstract reasoning. We show the published results as display-only evidence, excluded from overall and category rankings.

### Which model leads the published Humanity's Sixth Sense (HSS) snapshot?

Gemini-4-Argon (high) currently leads the published Humanity's Sixth Sense (HSS) snapshot with a score of 55.40%.

### How many models are evaluated on Humanity's Sixth Sense (HSS)?

The October 15, 2026 snapshot contains 27 AI models.

### Does Humanity's Sixth Sense (HSS) affect BenchLM's overall score?

Not directly. Humanity's Sixth Sense (HSS) is still displayed on BenchLM for reference, but it is excluded from the weighted scoring formula.