Atlas

Benchmarks

← All benchmarks

SMT 2025 (CAISI 58-task set)

Math · 2025-09-30

CAISI's SMT 2025 set contains 58 text-only advanced high-school mathematics problems from the Stanford Math Tournament. It is distinct from Atlas's 53-task MathArena track, and scores report answer accuracy.

Top models (higher is better)

ModelScore
GPT-5.599.0
GLM-5.297.8
DeepSeek-V4-Pro96.0
Opus 4.895.7
Claude Mythos Preview95.0
Opus 4.694.0
GPT-5.4 Mini92.0
GPT-591.8
DeepSeek-R1-052887.6
DeepSeek-V3.186.2
gpt-oss-120b82.3
Opus 482.2
DeepSeek-R175.0
Loading Atlas data…