GPT-5.6 Achieves IQ Score of 136, Surpassing 99% of Humans

The latest iteration of OpenAI's language model, GPT-5.6, has reportedly achieved an intelligence quotient (IQ) of 136 in offline tests conducted by Tracking AI. This score breaks the 130-point benchmark, widely considered the threshold for genius-level intellect in humans. The results suggest that GPT-5.6 now surpasses approximately 99% of the human population in cognitive tests.

Tracking AI utilized a proprietary, non-public dataset for its evaluation, designed to prevent models from simply recalling pre-learned answers. This stringent testing methodology saw multiple versions of GPT-5.6, including its visual counterparts, all scoring 136. This performance places it ahead of other leading models such as Claude-5 Fable, which scored 130, and previous GPT versions like GPT-5.6 LUNA Max that scored between 117 and 123.

While the 136 score is notable, AI researchers caution that IQ tests primarily measure abstract pattern recognition and logical reasoning. These tests do not fully encompass the breadth of intelligence, such as practical problem-solving skills, real-world application, or tool-use capabilities. The GPT-5.6 model's performance on these standardized mental tests is only one facet of its overall development.

Despite the limitations of IQ metrics, anecdotal evidence from developers suggests practical improvements. Users have reported GPT-5.6's enhanced ability to generate complex applications, debug code efficiently, and produce functional outputs from simple prompts. This suggests a growing alignment between theoretical cognitive scores and practical utility, although the path towards artificial general intelligence (AGI) remains a subject of ongoing research and debate.