Top 10 AI Models by IQ Score, 2025
Summary
The data measures IQ scores of top 10 text-only AI models on the Mensa Norway test for 2025. OpenAI o3 tops the list at 135, which is genius level. All models score above average human IQ of 90-110. This shows AI now matches or beats high human intelligence in text reasoning.
Key findings
- OpenAI o3 achieves top IQ score of 135.
- Claude-4 Sonnet scores 127 in second place.
- Gemini 2.0 Flash reaches 126 IQ.
- All 10 models exceed human average of 110.
- OpenAI o1 Pro lowest at 102 IQ.
Metrics Framework
- IQ score from Mensa Norway test measures intelligence.
- Unit is IQ points; above 130 indicates genius level.
- Data covers top 10 text-only AI models for 2025.
- Test is difficult exam for human intelligence evaluation.
Tags
Table View
| Model | Iq Score |
|---|---|
| OpenAI o3 | 135 |
| Claude-4 Sonnet | 127 |
| Gemini 2.0 Flash Thinking Exp. | 126 |
| Gemini 2.5 Pro Exp. | 124 |
| OpenAI o4 mini | 122 |
| Claude-4 Opus | 120 |
| Grok-3 Think | 112 |
| DeepSeek R1 | 106 |
| Llama 4 Maverick | 105 |
| OpenAI o1 Pro | 102 |
Analysis
Related Visuals
Top 10 AI Models by Performance Score, 2026
Top 10 AI Models LMSYS Text Elo, 2026
Top 5 AI Models by LiveBench Score, 2026
FAQ
IQ score comes from Mensa Norway test. It evaluates intelligence through difficult problems. The test is used for humans and applied to text-only AI models here. Scores show reasoning ability. Average human range is 90-110. Data lists exact scores for top 10 models.
Score above 130 means genius level. OpenAI o3 at 135 qualifies. Scores over 120 like 127 for Claude-4 Sonnet show strong performance. All top 10 beat human average 90-110. High scores indicate better text reasoning on Mensa test.
Data ranks top 10 AI models by IQ for 2025. It uses Mensa Norway test results. Scores reflect model performance in that year. No earlier years given. Focus is text-only models only.
Ranking uses Mensa Norway test for text reasoning. Vision models score lower, like 63 for GPT-4o Vision. Only top 10 text-only models shown. Test focuses on specific intelligence areas. No other benchmarks included in this data.