gpt-5.5 hits iq 136: new benchmark ranks 40+ ai models against humans

new benchmark shows why you better pay for ai than for human workforce

on a standard human iq bell curve, 50% of people score below 100. only 15% reach above 115 – the threshold typically associated with professional and academic high achievers. and above 130? that's roughly 2% of the population

ai iq, a benchmark site that launched 2 days ago, plotted 40+ of today's most popular ai models on that same curve. the numbers don't flatter us:

• only 20% of models score below 85 – mostly older or smaller models like gpt-4-turbo, opus-3, and grok-2
• 17% sit in the human-average zone of 85-100
• 32% land in the 100-115 range – above average by human standards
• roughly 30% score above 115, a tier only 15% of humans ever reach

and then there's the top of the leaderboard, where things get genuinely uncomfortable to look at:

1. gpt-5.5 – 136 iq
2. gpt-5.4 – 134 iq
3. gemini-3.1-pro – 132 iq
3. opus-4.7 – 132 iq
4. opus-4.6 – 131 iq
5. gemini-3-pro – 126 iq
5. gpt-5.2 – 126 iq
6. grok-4.3 – 125 iq
7. grok-4.20 – 123 iq
7. opus-4.5 – 123 iq
8. kimi-k2.6 – 122 iq
9. gpt-5.1 – 120 iq
10. gpt-5 – 119 iq
11. kimi-k2.5 – 118 iq
12. muse-spark – 117 iq
12. deepseek-v4-pro – 117 iq
13. glm-5.1 – 115 iq
14. minimax-m2.7 – 114 iq

a few things jump out:

• openai is absolutely dominating the top – gpt-5.5 and gpt-5.4 sit at 136 and 134, and they have four more models in the top 16. no other lab comes close in terms of sheer volume at the top
• anthropic and google are in a dead heat for second place – opus-4.7 and gemini-3.1-pro both land at 132, with opus-4.6 just behind at 131 and gemini-3-pro at 126
• xai is punching above its weight – grok-4.3 at 125 and grok-4.20 at 123 put elon's lab ahead of most of the field, which wasn't the narrative a year ago
• the chinese labs are closing the gap fast – kimi-k2.6 at 122, deepseek-v4-pro at 117, glm-5.1 at 115, minimax-m2.7 at 114. none of them are at the very top yet, but they're clearly no longer trailing by a wide margin
• muse-spark is a surprise – meta's model at 117 is genuinely competitive, quietly sitting above several well-known western models
• the gap between first and last on this list is only 22 iq points – which sounds small until you remember that on a human scale, 22 points is the difference between average and near-genius

for context, an iq of 130 in humans means you're smarter than 98% of the population. these models aren't just nudging past that line – they're stacking dimension scores well above it across the board. and unlike a human genius, they're available via api, right now, at a few dollars per task

it's already hard to compete with ai on raw intelligence. it knows more, reasons faster, and doesn't have bad days. but look at the trajectory – the frontier has moved from roughly 100 to 136 in under two years. if that pace continues, the gap won't just be uncomfortable. it will be impossible to close