벤치마크

뉴스에서 언급된 벤치마크 — 화제성 높은 순.

# 벤치마크 유형 선두 모델 언급
01 Terminal-Bench agentic 86.5% LLM-as-a-Verifier 4 4
02 OSWorld agentic 62.6% GPT-5.6 Sol 2 2
03 PutnamBench math 587 Leanstral 1.5 1 2
04 SWE-rebench agentic 56.5% Claude Opus 4.8 xhigh 10 1
05 ARC-AGI 2 reasoning 72.9% solver 3 1
06 JailbreakBench safety 93% STEER (applied to six open-source 8B-parameter models) 2 1
07 MATH-500 math 87% Qwen3.6-27B 2 1
08 AIME 2024 math 62.4% Qwen3-1.7B 1 1
09 AIME 2025 math 50.8% DASH 1 1
10 BrowseComp agentic 92.2% GPT-5.6 Sol 1 1