CogToM

Tong et al., 2026

What it measures

A published test we don't mirror: it supplies a reference figure, not test items

CogToM tests theory of mind, working out what others believe, want and feel, across 46 paradigms from developmental psychology, well beyond the classic false-belief task.

It has over 8,000 items in Chinese and English, validated by 49 human annotators, and was run on 22 models including GPT-5.1 and Qwen3-Max.

Example question

Written for this page in the style of the test, not one of its questions, so the test set stays unspoiled

The cake in the cupboardspecimen · not scored

Read the story, then answer.

(a) Story

Ana and Ben put a cake in the fridge. Ana leaves the kitchen. Ben moves the cake to the cupboard. Through the window Ana sees him do it, but Ben doesn't notice her.

(b) Question

Where does Ben think Ana will look for the cake?

(a) In the fridge
(b) In the cupboard

Figure 1. A second-order false-belief item: what one person thinks another believes, a classic theory-of-mind paradigm. In the style of CogToM. Answer: (a). Ana knows the cake moved, but Ben doesn't know she saw.

Human vs AI

The faculty-level figures we take from it, 0–100

Social cognitionGap +4.4

Humans

84.4

Social IQa (Sap et al., 2019)

From another test: Social IQa

Best published AI

80.0

CogToM, GPT-5.1 (a different test)

These are third-party results on different questions from ours. Where the numbers come from

All reference tests →