Ryan Shea

Experiments in AI & bio: Former Senior Advisor, FDA. Cofounder, @Stacks. 10+ unicorn investor. Princeton MechE/CS. NJ Math League State Champ.

[Ryan Shea @ryaneshea]

Today I’m launching AI IQ — frontier AI models, scored on the human IQ scale.

Instead of endless leaderboard tables, AI IQ shows:

• Where models land on the IQ bell curve
• How frontier IQ is changing over time
• How models compare on IQ and EQ
• What intelligence costs in practice

GPT-5.5, Claude Opus 4.7, Gemini 3.1, Grok 4.3, Kimi K2.6, Qwen3.6, DeepSeek V4, Muse Spark, and more.

Link in the first reply. Curious which chart surprises you most.

[Ryan Shea @ryaneshea]

@BenDiFrancesco The tweet I'm quote-tweeting is a joke that was cleverly crafted to illustrate a real concern that could materialize.

[Ryan Shea @ryaneshea]

Nightmare scenario of US AI models refusing to help w/ frontier LLM research btw:

Chinese OS models become SOTA for LLM research and shift devs worldwide to Chinese AI accelerators.

This would cede the US compute advantage.

US hardware must win.

Jensen was right.

[kache @yacineMTB]

That's weird.. GLM 5.2 is refusing to write CUDA kernels for my b300. But, thankfully, it is happily optimizing the kernels for my Huawei ascends! And they're much cheaper anyways..

[Ryan Shea @ryaneshea]

@chamath GLM-5.2 is at the frontier across other frontend benchmarks as well so it's not simply due to benchmaxxing / gaming a single benchmark:
https://t.co/DlMwJzLwNJ

[Ryan Shea @ryaneshea]

GLM-5.2 is ranked #2 on Arena for Frontend Engineering

You can locally run a model that is better than Opus

You can locally run a model that is nearly as good as Mythos for 10% of the cost

Absolutely insane

https://t.co/apfvt6DZX5

[Ryan Shea @ryaneshea]

A single benchmark delivers limited signal of course so I'm basing this conclusion off of multiple benchmarks and the result translates across them.

[Ryan Shea @ryaneshea]

Whoa, this is absolutely insane:

GLM-5.2 beats Fable 5 on overall coding ability on Design Arena

I don't think people appreciate how big of a deal this is

[Anthony @anthonystaj]

so fable level model is now free? @ all of biox

[Ryan Shea @ryaneshea]

https://t.co/pIqCejr8A6

[Ryan Shea @ryaneshea]

This has never happened before.

Open source models have never matched frontier models on coding ability. Ever.

For the first time, open source has caught up with the closed source frontier.

[Ryan Shea @ryaneshea]

This has never happened before.

Open source models have never matched frontier models on coding ability. Ever.

For the first time, open source has caught up with the closed source frontier.

[Ryan Shea @ryaneshea]

GLM-5.2 is ranked #2 on Arena for Frontend Engineering

You can locally run a model that is better than Opus

You can locally run a model that is nearly as good as Mythos for 10% of the cost

Absolutely insane

https://t.co/apfvt6DZX5

[Ben Liu @BenjamineYLiu]

https://t.co/ExUEonRIQI

[Ryan Shea @ryaneshea]

@SinaHartung Yes, it's absolutely night and day.

[Ryan Shea @ryaneshea]

@SalsaTekila Fable 5's IQ is about 131

Opus 4.8's IQ is about 128

[Daniel @growing_daniel]

Kinda weird we post on spacex

[Ryan Shea @ryaneshea]

@Ronalfa @NOETIK_ai Awesome work Ron, you guys are crushing it!

[Ryan Shea @ryaneshea]

@BrendanFoody For what model to use for various tasks, you should check out AI IQ.

Best collection of evals in the industry.

[Ryan Shea @ryaneshea]

@DillonUzar @AnthropicAI @OpenRouter Great benchmark! Can you add more models?

[Ryan Shea @ryaneshea]

@tszzl @tamaybes Becoming clear to me that we haven’t invented enough words for ourselves to think in high enough dimensions and levels of detail

[Ryan Shea @ryaneshea]

@LechMazur Excellent benchmarks!

[Ryan Shea @ryaneshea]

@tszzl Risks regarding entities that can replicate and take up residence in infected hosts, digital or biological

[Ryan Shea @ryaneshea]

@SynBio1 @antonioregalado @purrmin I’ll bet on many forms of market expansion:

1. Total US drug spending goes higher
2. Trials get faster and cheaper, increasing margins or decreasing the minimum price charged
3. Population increases
4. New drugs displace old drugs at a higher rate than before

[Ryan Shea @ryaneshea]

@OfirPress SWE-Bench Verified and CritPt are two of the benchmarks we track on AI IQ:

https://t.co/WLDDRJj3Qi

[Ryan Shea @ryaneshea]

@OfirPress You produce fantastic benchmarks sir! CritPt is goated.