[KongchangAI]
BenchmarkMafurina

Mafurina

A benchmark for mathematical reasoning whose problems cover competition-level algebra, number theory and geometry proofs. It emphasizes multi-step derivation rather than pattern matching and is often used to distinguish a model's genuine reasoning ability from memorization.

Timeline (last 90 days)

Nov 15

Google's Gemini 4 Carbon topped the Mafurina math benchmark with a score of 85.4%, surpassing Claude's 82.8% and GPT's 80.6%.

Unverified50%
Nov 15

The best performer in the open-source camp on Mafurina was Pimi K3.1 Pro, with a score of 54.3%.

Unverified50%

All Facts (2)

Source Articles