List your company
ResearchFormerly WizWand

Math Problem Solving on 200 IMO-level math problems IMO-AnswerBench, IMO-ProofBench, ArXivMath (test)

50.6Pass@1 AccuracyMeta-Harness
21.89629.34836.844.252Mar 30, 2026
Updated 7mo ago

Evaluation results

MethodLinks
Meta-HarnessBackbone Model=GPT-20B, Retrieval Strategy=Meta-Harness2026.0350.6
BM25 RetrievalBackbone Model=GPT-20B, Retrieval Strategy=BM25 Retrieval2026.0348.9
No RetrieverBackbone Model=GPT-20B, Retrieval Strategy=None2026.0347.6
Dense RetrievalBackbone Model=Gem-3F, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=52026.0347.2
Dense RetrievalBackbone Model=GPT-20B, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=12026.0346.9
Dense RetrievalBackbone Model=GPT-20B, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=52026.0346.7
BM25 RetrievalBackbone Model=Gem-3F, Retrieval Strategy=BM25 Retrieval2026.0346.6
Meta-HarnessBackbone Model=Gem-3F, Retrieval Strategy=Meta-Harness2026.0346.3
No RetrieverBackbone Model=Gem-3F, Retrieval Strategy=None2026.0342.6
Dense RetrievalBackbone Model=Gem-3F, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=12026.0342.3
Random Few-shotBackbone Model=GPT-20B, Retrieval Strategy=Random Few-shot2026.0341.8
Random Few-shotBackbone Model=Gem-3F, Retrieval Strategy=Random Few-shot2026.0340.4
Meta-HarnessBackbone Model=Avg., Retrieval Strategy=Meta-Harness2026.0338.8
Dense RetrievalBackbone Model=Avg., Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=52026.0338.1
BM25 RetrievalBackbone Model=Avg., Retrieval Strategy=BM25 Retrieval2026.0337.5
Dense RetrievalBackbone Model=Gem-3.1FL, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=52026.0337.1
Meta-HarnessBackbone Model=Gem-3.1FL, Retrieval Strategy=Meta-Harness2026.0334.9
Dense RetrievalBackbone Model=Avg., Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=12026.0334.4
No RetrieverBackbone Model=Avg., Retrieval Strategy=None2026.0334.1
BM25 RetrievalBackbone Model=Gem-3.1FL, Retrieval Strategy=BM25 Retrieval2026.0332.8
Random Few-shotBackbone Model=Avg., Retrieval Strategy=Random Few-shot2026.0332.2
Meta-HarnessBackbone Model=GPT-5.4n, Retrieval Strategy=Meta-Harness2026.0331.7
Dense RetrievalBackbone Model=Gem-3.1FL, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=12026.0331.3
Dense RetrievalBackbone Model=GPT-5.4n, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=52026.0331.1
Random Few-shotBackbone Model=Gem-3.1FL, Retrieval Strategy=Random Few-shot2026.0331
Meta-HarnessBackbone Model=GPT-5.4m, Retrieval Strategy=Meta-Harness2026.0330.4
BM25 RetrievalBackbone Model=GPT-5.4n, Retrieval Strategy=BM25 Retrieval2026.0330.2
BM25 RetrievalBackbone Model=GPT-5.4m, Retrieval Strategy=BM25 Retrieval2026.0329.2
No RetrieverBackbone Model=GPT-5.4m, Retrieval Strategy=None2026.0328.8
No RetrieverBackbone Model=Gem-3.1FL, Retrieval Strategy=None2026.0328.6
Dense RetrievalBackbone Model=GPT-5.4m, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=52026.0328.3
Dense RetrievalBackbone Model=GPT-5.4n, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=12026.0327.1
Dense RetrievalBackbone Model=GPT-5.4m, Retrieval Strategy=Dense Retrieval, k (Retrieval Count)=12026.0324.5
Random Few-shotBackbone Model=GPT-5.4m, Retrieval Strategy=Random Few-shot2026.0324.5
Random Few-shotBackbone Model=GPT-5.4n, Retrieval Strategy=Random Few-shot2026.0323.1
No RetrieverBackbone Model=GPT-5.4n, Retrieval Strategy=None2026.0323