AI IQ: Ranking Language Models Like Humans

A startup project named AI IQ has introduced a novel way to evaluate artificial intelligence by assigning estimated intelligence quotients (IQ) to over 50 of the world's most powerful language models. These scores are presented through interactive visualizations, which have garnered both praise for making the AI market more accessible and sharp criticism from researchers who deem the framework misleading.

The AI IQ methodology aggregates 12 benchmarks across four reasoning dimensions: abstract, mathematical, programmatic, and academic. The composite IQ is derived as a straight average of these dimension scores. Each benchmark score is mapped to an implied IQ using calibrated difficulty curves, with ceilings compressed for easier benchmarks to prevent inflated scores.

As of mid-May 2026, the AI IQ charts indicate a significant convergence among top AI models, with OpenAI's GPT-5.5 leading at an estimated IQ near 136, closely followed by models from Anthropic and Google. Below this frontier cluster, a competitive midfield emerges, with several Chinese AI models showing strong performance in the 112-118 IQ range.

Beyond cognitive intelligence, AI IQ also incorporates an "EQ" or emotional intelligence score, derived from EQ-Bench 3 and Arena Elo scores. This EQ ranking introduces a different perspective, with Anthropic's Opus 4.7 leading in EQ, suggesting a more balanced intelligence profile compared to some high-IQ-only models.

Perhaps the most practically relevant visualization for enterprise buyers is the IQ versus Effective Cost scatter plot. This chart highlights that while leading models like GPT-5.5 and Opus 4.7 are expensive, more economical options like GPT-5.4-mini and DeepSeek-V3.2 offer respectable IQ scores at significantly lower costs.

Critics, however, raise philosophical objections, arguing that a single IQ score dangerously oversimplifies the "jagged" and uneven capabilities of AI models. Concerns also exist regarding the transparency of calibration curves and the potential for bias, although AI IQ acknowledges some scoring biases and implements corrective measures.

Despite its imperfections, AI IQ provides a unified framework for comparing models across providers and price points, a significant improvement over wading through disparate, provider-specific benchmark tables. The project underscores the growing importance of "orchestration"—wisely selecting and deploying the right AI model for specific tasks—as a critical skill in the evolving AI landscape.