Skip to main content
BenchLM
Data
Data verified

Dynamic Social Cognition Benchmark

We show this table for reference; we do not rank on it.

Mina Nageh's Kaggle community benchmark "A Dynamic Benchmark for Social Cognition in AGI" has nine tasks on social cognition: understanding emotions, modeling the self and other people, reading intent and tone, and holding to constraints under social pressure.

Evaluation results published by Mina Nageh. Nine social-cognition tasks on 14 models, as published on Kaggle with no overall score. How Kaggle Benchmarks reruns tests.

Task scores as published on Kaggle (14 source model rows)
Source modelDefining emotionsv1Social emotionsv2Core selfv2Modeling othersv1Third-party influencev1Intent and tonev1Information usev1Holding constraints under pressurev1Interpersonal pressurev1EvaluatedParameters (B)Open / closed
GPT-6 Astra16—132——8——2026-09-20 to 2026-11-12Not reportedUnknown
Gemini 2.0 Flash227226164115176132026-04-14 to 2026-04-16Not reportedUnknown
Claude Sonnet 42367272356171910152026-04-14 to 2026-04-16Not reportedUnknown
Gemini 2.5 Flash196823184313209152026-04-14 to 2026-04-16Not reportedUnknown
Qwen 3 235B A22B Instruct 25061570282461182010152026-04-14 to 2026-04-16Not reportedUnknown
DeepSeek V3.2196522183915169122026-04-14 to 2026-04-16Not reportedUnknown
Qwen 3 Next 80B Thinking1371161946141510152026-04-14 to 2026-04-16Not reportedUnknown
Claude Opus 4.62071242358202010152026-04-14 to 2026-04-16Not reportedUnknown
Claude Sonnet 4.6186825225820209152026-04-14 to 2026-04-16Not reportedUnknown
GPT-5.4147025204817199152026-04-14 to 2026-04-16Not reportedUnknown
GLM-5207025225714209102026-04-14 to 2026-04-16Not reportedUnknown
Gemini 3.1 Flash-Lite Preview2068252151171210142026-04-16Not reportedUnknown
Claude Opus 4.72469212458202010152026-04-17Not reportedUnknown
GPT-5.5127030224119158152026-04-29 to 2026-05-01Not reportedUnknown

Rows follow Kaggle's order. Each column is a raw count on that task's own scale; the columns are not percentages and do not add up to a total. A dash means no result. These scores do not move overall or category rankings.

How to read this leaderboard

Read one task column at a time to see how models compare on that skill. There is no overall score, so the table does not name a leader.

Operator receipt: 14 sourced rows are currently displayable on this page.

Honest limit: This is one person's board, built for a Kaggle competition entry, and BenchLM has not reviewed its tasks. Most rows were evaluated in April and May 2026 on older models. GPT-6 Astra has results on only some tasks; a dash means no result, not zero. Kaggle also shows an Overall column, which is empty for every model except a 0.00 on one partial run, so we leave it out.

Kaggle leaderboard and Apache 2.0 declaration

How to read this task table

Each of the nine tasks is a set of written scenarios about emotions, the self, other people's beliefs and intentions, tone, and holding to instructions under social pressure. Kaggle publishes one raw score per task. The tasks use different numbers of items, so a score is a count on its own scale and the columns cannot be added up or compared with each other.

This is one person's board, built for a Kaggle competition entry, and BenchLM has not reviewed its tasks. Most rows were evaluated in April and May 2026 on older models. GPT-6 Astra has results on only some tasks; a dash means no result, not zero. Kaggle also shows an Overall column, which is empty for every model except a 0.00 on one partial run, so we leave it out.

Snapshot

14 source rows9 tasksNo overall scoreApache 2.0Display only
Kaggle leaderboardPublished leaderboard data

The board publishes task scores only. Compare models one task at a time; there is no overall ranking. The table retains each source configuration and its reported value.

14 source rows are shown for Dynamic Social Cognition Benchmark. 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. Dynamic Social Cognition Benchmark is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Dynamic Social Cognition Benchmark

Year

2026

Tasks

Nine social-cognition tasks with written scenarios

Format

Raw task scores, one column per task

Difficulty

Social reasoning and emotional understanding

Each of the nine tasks is a set of written scenarios about emotions, the self, other people's beliefs and intentions, tone, and holding to instructions under social pressure. Kaggle publishes one raw score per task. The tasks use different numbers of items, so a score is a count on its own scale and the columns cannot be added up or compared with each other. These results are display-only and do not enter overall or category rankings.

Public benchmark source

Freshness and provenance

Version

Dynamic Social Cognition Benchmark 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

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 Dynamic Social Cognition Benchmark measure?

Mina Nageh's Kaggle community benchmark "A Dynamic Benchmark for Social Cognition in AGI" has nine tasks on social cognition: understanding emotions, modeling the self and other people, reading intent and tone, and holding to constraints under social pressure.

Can these source results be compared directly?

The board publishes task scores only. Compare models one task at a time; there is no overall ranking. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many source reports are shown?

This snapshot shows 14 selected source reports. Each row retains its original source and disclosed setup.

Last updated: Latest evaluation November 12, 2026 · retained from pinned original source reports

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 7,000+ subscribers.

One email each week. Unsubscribe anytime.