Skip to main content
MANIFOLD
When will WeirdML be at ≥ 80% official score?
1
Ṁ100Ṁ25
2027
34%
Before November 2026
41%
Before January 2027
50%
Before March 2027
66%
Before May 2027

WierdML (https://htihle.github.io/weirdml.html) just got it's third release. How fast will LLMs saturate this benchmark? The current best model is GPT-6 Astra with 42%. Note that the "official score" is different from the "final best" as it also takes the token efficiency into account by using the area under the curve.

The models has to be publicly released to count.

If the benchmark is rewored, similar to FrontierMath, then the reworked scores count.

I will add more buckets as needed.

Get
Ṁ1,000
to start trading!