Rabdos AI

Rabdos Math Bench

For research-level mathematical reasoning.

Leaderboard

36 problems · Updated Jul 31, 2026

Leaderboard, ranked by average score.
RankProviderModelReasoning effortGraded score
1OpenAIGPT-5.6 Solmax41.7%
1AnthropicClaude Fable 5max41.7%
3AnthropicClaude Opus 4.8max38.9%
4MetaMuse Spark 1.1xhigh30.6%
5Moonshot AIKimi K3max27.8%
6xAIGrok 4.5xhigh27.8%
7GoogleGemini 3.6 Flashhigh25.0%
8Z.aiGLM 5.2xhigh19.4%
9DeepSeekDeepSeek V4 Proxhigh16.7%
10NVIDIANemotron 3 Ultrahigh16.7%

Sample problems

Shadows of uniform set families

Extremal set theory

For a finite family F\mathcal{F} of distinct kk-element sets, let d(F)d(\mathcal{F}) be the collection of all (k1)(k-1)-element sets contained in at least one member of F\mathcal{F}, and write D(F)=d(F)D(\mathcal{F}) = |d(\mathcal{F})|. Let M(k)M(k) be the largest positive integer that is not equal to D(F)D(\mathcal{F}) for any finite family F\mathcal{F} of distinct kk-element sets. For fixed kk and tt, let Lk(t)L_k(t) be the smallest integer LL such that every integer from LL through tktk is equal to D(F)D(\mathcal{F}) for some tt-member family F\mathcal{F} of distinct kk-element sets. Let τ(k)\tau(k) be the largest tt in {1,,k+1}\{1, \ldots, k+1\} with Lk(t)tkk+2L_k(t) \ge tk - k + 2.

Low-stem squares in the cobar complex

Algebraic topology

Work over F2\mathbb F_2. Let A=F2[ξ1,ξ2,]A=\mathbb F_2[\xi_1,\xi_2,\ldots], with ξi=2i1|\xi_i|=2^i-1 and coproduct

Δ(ξn)=i=0nξni2iξi,\Delta(\xi_n)=\sum_{i=0}^{n}\xi_{n-i}^{\,2^i}\otimes\xi_i,

Crossings of polygonal curves

Discrete geometry

A closed polygonal curve in the plane is called generic if its vertices are distinct, no three vertices are collinear, no vertex lies in the relative interior of a nonincident edge, and no three edges have a common point. Thus every intersection between nonadjacent edges is a transverse double crossing.

Interested in evaluating your model on Rabdos Math Bench? Get in touch →