Human Behavior Bench V1 (Kaggle) (Human Behavior Bench)
We show this table for reference; we do not rank on it.
Predicts response distributions for specified groups of U.S. adults on closed-form questions about behavior and preferences.
Evaluation results published by Kaggle Benchmarks. How Kaggle Benchmarks reruns tests.
Distribution agreement on Human Behavior Bench — 2026-10-20 source capture
We mirror the published distribution agreement view for Human Behavior Bench. Gemini 4 Argon has the highest reported value at 80.6%. These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. We do not use these results to rank models overall.
Gemini 4 Argon
Task version 1; Evaluated 2026-10-13
Claude Opus 5
Task version 1; Evaluated 2026-09-30
Gemini 3.7 Flash
Task version 1; Evaluated 2026-10-02
11 configurationsReasoningCurrentDisplay onlyUpdated 2026-10-20 source capture
Distribution agreement table (11 configurations)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Gemini 4 ArgonCreator not reportedTask version 1; Evaluated 2026-10-13 | 80.6% | Not reported | Unknown |
| 2 | Claude Opus 5Creator not reportedTask version 1; Evaluated 2026-09-30 | 80.2% | Not reported | Unknown |
| 3 | Gemini 3.7 FlashCreator not reportedTask version 1; Evaluated 2026-10-02 | 79.7% | Not reported | Unknown |
| 4 | GPT-5.6 SolCreator not reportedTask version 1; Evaluated 2026-10-05 | 79.5% | Not reported | Unknown |
| 5 | Claude Opus 5.5Creator not reportedTask version 1; Evaluated 2026-10-05 | 79.3% | Not reported | Unknown |
| 6 | Gemini 3.8 FlashCreator not reportedTask version 1; Evaluated 2026-10-05 | 79.2% | Not reported | Unknown |
| 7 | Gemini 3 Flash PreviewCreator not reportedTask version 1; Evaluated 2026-10-02 | 78.5% | Not reported | Unknown |
| 8 | GPT-5.6 TerraCreator not reportedTask version 1; Evaluated 2026-09-30 | 77.8% | Not reported | Unknown |
| 9 | Gemini 3.5 Flash-LiteCreator not reportedTask version 1; Evaluated 2026-09-30 | 75.9% | Not reported | Unknown |
| 10 | Grok 4.7Creator not reportedTask version 1; Evaluated 2026-10-16 | 75.2% | Not reported | Unknown |
| 11 | Grok 4.20 (Non-Reasoning)Creator not reportedTask version 1; Evaluated 2026-10-05 | 73.5% | Not reported | Unknown |
How to read this leaderboard
Operator receipt: 11 sourced rows are currently displayable on this page; the leading published row is Gemini 4 Argon at 80.6%.
Honest limit: The evaluation concerns group survey responses in the United States. It does not establish individual behavior, future purchases, or performance in other countries. The owner and Kaggle descriptions report different participant totals, so we do not combine those figures.
How to read these results
Human Behavior Bench asks a model to predict how a specified group of U.S. adults answered a closed-form survey question. The V1 dataset has 1,199 tasks: 239 public and 960 private.
We show the Kaggle Combined Score as a percentage. Agreement is one minus total variation distance between the predicted and observed answer distributions. Public and private scores remain separate in the source snapshot.
This board contains Kaggle model evaluations. Rehearsals Engram uses a separate in-house simulation harness on the owner website and is not merged into this table. Survey agreement does not establish individual behavior or future purchasing accuracy.
Snapshot
These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. The table retains each source configuration and its reported value.
11 configurations are shown for Human Behavior Bench. The benchmark falls in the Reasoning category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Human Behavior Bench is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Human Behavior Bench
Year
2026
Tasks
1,199 tasks; 239 public and 960 private
Format
Published Kaggle leaderboard aggregates
The combined agreement score is one minus total variation distance, shown on a 0–100 percentage scale. The public and private task splits remain available in the source snapshot. We keep the Kaggle harness separate from the owner's Engram simulation report.
Freshness and provenance
Version
Human Behavior Bench 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Human Behavior Bench measure?
Predicts response distributions for specified groups of U.S. adults on closed-form questions about behavior and preferences.
Which source row has the highest reported Human Behavior Bench value?
Gemini 4 Argon has the highest reported Human Behavior Bench value at 80.6%. These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Human Behavior Bench?
The 2026-10-20 source capture snapshot contains 11 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.