Human attention ESTIMATED
Between people · 01
78.00
90% interval 47.27–97.52 · P(rarest) 36%
GZ—
MI—
GX—
BB—
Human Skills Leaderboard — October 2026
As of October 14, 2026, BenchHuman shows its research starting point for 9 human skills. Community votes replace it from Snapshot #1. Human attention and Reflection tie under the 90% interval rule.
Snapshot #0 · research priorMethod v2.0.0 · Beta-Binomial + Bradley-TerryCounting new votes…
AT
Reflection ESTIMATED
Inner life · 02
72.00
90% interval 39.78–95.61 · P(rarest) 21%
GZ—
MI—
GX—
BB—
Listening to people ESTIMATED
Between people · 03
66.00
90% interval 33.83–92.31 · P(rarest) 12%
GZ—
MI—
GX—
BB—
Participants0
Ratings0
Arena battles0
Snapshot dateOctober 14, 2026
The human skills leaderboard
Score = estimated % of people who say a skill is harder to develop today than five years ago. Arena Elo comes from head-to-head votes. Generation columns show the same score per age group once 3 people from that group have voted.
SUPPORTED 30+ votesESTIMATED fewer votes, wider intervalGZ Gen Z · MI Millennials · GX Gen X · BB Boomers+
| Skill | Score ⌄ | 90% interval | Arena Elo | P(rarest) | Votes | GZ | MI | GX | BB | Key evidence | |
|---|---|---|---|---|---|---|---|---|---|---|---|
| #1 | AT Human attention Between people · ESTIMATED | 78.00 | 47.27–97.52 | 1500 | 36% | 0 | — | — | — | — | Cognitive Divergence, arXiv 2603.26707 (2026) |
| #2 | RF Reflection Inner life · ESTIMATED | 72.00 | 39.78–95.61 | 1500 | 21% | 0 | — | — | — | — | Gerlich, Societies 2025 (n=666) |
| #3 | LI Listening to people Between people · ESTIMATED | 66.00 | 33.83–92.31 | 1500 | 12% | 0 | — | — | — | — | WEF Future of Jobs 2025 |
| #4 | AU Authenticity Inner life · ESTIMATED | 64.00 | 32.01–90.93 | 1500 | 10% | 0 | — | — | — | — | Editorial prior |
| #5 | ID Identity Inner life · ESTIMATED | 60.00 | 26.86–89.25 | 1500 | 8% | 0 | — | — | — | — | arXiv 2603.29886 (2026) |
| #6 | EM Empathy Between people · ESTIMATED | 58.00 | 26.04–86.95 | 1500 | 5% | 0 | — | — | — | — | WEF Future of Jobs 2025 |
| #7 | PH Philosophy Inner life · ESTIMATED | 55.00 | 23.06–84.96 | 1500 | 4% | 0 | — | — | — | — | Editorial prior |
| #8 | FE Bringing people together Together · ESTIMATED | 52.00 | 21.20–83.06 | 1500 | 3% | 0 | — | — | — | — | Editorial prior |
| #9 | LE Leadership Together · ESTIMATED | 45.00 | 15.37–77.62 | 1500 | 1% | 0 | — | — | — | — | WEF Future of Jobs 2025 |
| Below: today's top AI models on BenchLM (October 10, 2026). Strong on agentic, coding and reasoning tests. They cannot practise these nine skills for you. | |||||||||||
| — | AI Claude Opus 5.5 Anthropic | n/a | BenchLM score 86.43. Human-skill score not applicable. | ||||||||
| — | AI GPT-6 Astra OpenAI | n/a | BenchLM score 84.88. Human-skill score not applicable. | ||||||||
| — | AI GPT-6.1 Sol OpenAI | n/a | BenchLM score 84.18. Human-skill score not applicable. | ||||||||
Vote
Two ways to vote, no account needed. The arena is quick: pick the harder one, as often as you like. The full rating feeds the score and the generation columns.
Arena · which is harder to develop today?
VS
Pick the one you find harder to develop today.
Full rating · compared to 5 years ago
1 much easier3 same5 much harder
Human attention
Reflection
Listening to people
Authenticity
Identity
Empathy
Philosophy
Bringing people together
Leadership
One rating per person. Submitting again updates it.
How the score is built
Every snapshot is computed from raw events with a fixed seed, so anyone can recompute it from the open data. Methods follow LMArena's Arena-Rank (Bradley-Terry) and open opinion-mapping practice.
Score · Beta-Binomial
A rating of 4 or 5 counts as "harder". Each skill starts from a weak research prior worth 6 pseudo-votes. Real votes outweigh it quickly. Unverified votes count half.
θ ~ Beta(6·p₀ + k, 6·(1−p₀) + n − k)
Arena Elo · Bradley-Terry
Every head-to-head vote is a match. Ratings are fitted on the whole history at once, so vote order does not matter, then put on an Elo scale centred at 1500.
P(i harder than j) = πᵢ / (πᵢ + πⱼ)
Ties and P(rarest)
4,000 posterior draws give the 90% interval and the chance each skill is truly #1. Neighbours whose 90% intervals overlap are reported as tied.
tie(a,b) ⇔ [lo_a, hi_a] ∩ [lo_b, hi_b] ≠ ∅
Priors are an editorial starting point, not survey results. Votes come from whoever takes part, so they are not a representative sample of any population. Open data: /api/v1/leaderboard (CC BY 4.0) · MCP server: https://benchhuman.com/api/mcp
The way back is a rhizome, not a ladder
Node size = betweenness centrality: how many shortest paths run through a skill. Colour = cluster. There is no root: enter anywhere, and each link you strengthen feeds the others. Skills people rate together add new links.
Influence hubs
The top hub is the cheapest place to start: it sits on the most paths.
#1
Empathy
0.308#2
Reflection
0.199#3
Listening to people
0.199#4
Doubt
0.127#5
Identity
0.109#6
Bringing people together
0.095Structural gap: Inner life ↔ Together
Knowing yourself does not reach the group on its own. The dashed practices below bridge that gap.
Human attention ⟷ Listening to people
Ten minutes, no input
Once a day, sit with nothing to read, watch or ask. The attention you rebuild alone is the attention you bring to the next person.
Philosophy ⟷ Listening to people
Ask one real question
In every conversation, ask one question you don't know the answer to. Then listen to the whole answer.
Identity ⟷ Bringing people together
Tell your own story
Share where you come from and what you care about, unedited. People gather around stories, not summaries.
Authenticity ⟷ Leadership
Disagree kindly, out loud
Say what you actually think, with care. Leadership starts where people trust that you mean it.
Reflection ⟷ Empathy
Write before you prompt
Draft your own view first, then compare with the machine. You keep your thinking and gain a second perspective.
Empathy ⟷ Bringing people together
Host something offline
A dinner, a walk, a reading circle. Empathy grows into community when it has a place to meet.
Research radar
Source-linked studies behind the priors, newest first.
AI empathy and the cognitive autonomy of younger users
Mar 2026arXiv 2603.29886
The Cognitive Divergence: AI context windows, human attention decline and the delegation loop
Mar 2026arXiv 2603.26707
More frequent AI use linked to weaker critical thinking through cognitive offloading
Mar 2025Gerlich, Societies
Future of Jobs Report 2025: analytical thinking, resilience and leadership lead core skills
Jan 2025World Economic Forum