Atlas

Benchmarks

← All benchmarks

ArXivMath 12/2025

Math · 2026-02-04

This MathArena track evaluates models on the ArXivMath 12/2025 problem set. Scores report the percentage of problems answered correctly.

Top models (higher is better)

ModelScore
Gemini 3.1 Pro Preview66.2
GPT-5.460.3
Opus 4.657.4
GPT-5.252.2
Gemini 3 Pro Preview51.1
Grok 4.1 Fast50.0
Gemini 3 Flash Preview43.4
Kimi K2.541.9
Step 3.5 Flash41.9
DeepSeek-V3.241.5
Qwen3.5-27B41.2
Qwen3.5 35B A3B39.7
Qwen3.5-9B39.7
GLM-538.2
Qwen3.5 397B A17B38.2
Qwen3.5-4B34.1
Nemotron 3 Super 120B A12B33.8
Qwen3 4B Thinking 250732.4
Qwen3 30B A3B Thinking 250727.9
QED-Nano26.5
Loading Atlas data…