AetherCode厂商自报 · 越高越好
38.2%
Official Table 1; think-high; no tools; Pass@1
成绩出处AI2D厂商自报 · 越高越好
89.1%
Official Table 2; think-high; Pass@1
成绩出处AIME 2025厂商自报 · 越高越好
94.3%
Official Table 1; think-high; no tools; Pass@1
成绩出处AMO Bench厂商自报 · 越高越好
60%
Official Table 1; think-high; no tools; Pass@1
成绩出处AndroidWorld厂商自报 · 越高越好
70.7%
Official Table 5; think-high; agentic GUI evaluation
成绩出处ARC-AGI厂商自报 · 越高越好
67.9%
Official Table 1; think-high; no tools; Pass@1
成绩出处Beyond AIME厂商自报 · 越高越好
77%
Official Table 1; think-high; no tools; Pass@1
成绩出处BFCL-V4厂商自报 · 越高越好
57.2%
Official Table 4; think-high; benchmark-specific agent tools
成绩出处BLINK厂商自报 · 越高越好
74.3%
Official Table 2; think-high; Pass@1
成绩出处BrowseComp厂商自报 · 越高越好
67.6%
Official Table 4; think-high; benchmark-specific agent tools
成绩出处BrowseComp-zh厂商自报 · 越高越好
81.3%
Official Table 4; think-high; benchmark-specific agent tools
成绩出处CharXiv-R厂商自报 · 越高越好
71.4%
Official Table 2; think-high; Pass@1
成绩出处CountBench厂商自报 · 越高越好
96.3%
Official Table 2; think-high; Pass@1
成绩出处DUDE厂商自报 · 越高越好
69.4%
Official Table 2; think-high; Pass@1
成绩出处DynaMath厂商自报 · 越高越好
61.5%
Official Table 2; think-high; Pass@1
成绩出处EMMA厂商自报 · 越高越好
60.9%
Official Table 2; think-high; Pass@1
成绩出处ERQA厂商自报 · 越高越好
58.8%
Official Table 2; think-high; Pass@1
成绩出处Hallusion Bench厂商自报 · 越高越好
63.9%
Official Table 2; think-high; Pass@1
成绩出处Humanity's Last Exam (with tools, text-only)厂商自报 · 越高越好
40.9%
Official Table 4; think-high; agentic search; text-only with tools
成绩出处LiveCodeBench Pro厂商自报 · 越高越好
1,930分
Official Table 1; think-high; no tools; Elo rating
成绩出处LiveCodeBench v6厂商自报 · 越高越好
79.5%
Official Table 1; think-high; no tools; Pass@1
成绩出处LiveSports-3K厂商自报 · 越高越好
77.5%
Official Table 3; think-high; public video evaluation
成绩出处LongVideoBench厂商自报 · 越高越好
77.4%
Official Table 3; think-high; public video evaluation
成绩出处LVBench厂商自报 · 越高越好
73%
Official Table 3; think-high; public video evaluation
成绩出处MathVision厂商自报 · 越高越好
81.3%
Official Table 2; think-high; Pass@1
成绩出处MathVista厂商自报 · 越高越好
87.7%
Official Table 2; think-high; Pass@1
成绩出处Minerva厂商自报 · 越高越好
62.4%
Official Table 3; think-high; public video evaluation
成绩出处MMLU厂商自报 · 越高越好
92.3%
Official Table 1; think-high; no tools; Pass@1
成绩出处MMLU-Pro厂商自报 · 越高越好
84.9%
Official Table 1; think-high; no tools; Pass@1
成绩出处MMMU厂商自报 · 越高越好
83.4%
Official Table 2; think-high; Pass@1
成绩出处MMMU-Pro厂商自报 · 越高越好
73.2%
Official Table 2; think-high; Pass@1
成绩出处MMStar厂商自报 · 越高越好
79.9%
Official Table 2; think-high; Pass@1
成绩出处MMVU厂商自报 · 越高越好
73.1%
Official Table 3; think-high; public video evaluation
成绩出处MotionBench厂商自报 · 越高越好
70.6%
Official Table 3; think-high; public video evaluation
成绩出处MuirBench厂商自报 · 越高越好
78.7%
Official Table 2; think-high; Pass@1
成绩出处Multi-Challenge厂商自报 · 越高越好
66.7%
Official Table 1; think-high; no tools; Pass@1
成绩出处Multi-SWE-Bench厂商自报 · 越高越好
42%
Official Table 4; think-high; benchmark-specific agent tools
成绩出处OmniDocBench 1.5厂商自报 · 越高越好
10.6%
Official Table 2; think-high; normalized edit distance (lower is better)
成绩出处OSWorld厂商自报 · 越高越好
61.9%
Official Table 5; think-high; agentic GUI evaluation
成绩出处OVBench厂商自报 · 越高越好
65.1%
Official Table 3; think-high; public video evaluation
成绩出处OVOBench厂商自报 · 越高越好
72.6%
Official Table 3; think-high; public video evaluation
成绩出处PHYBench厂商自报 · 越高越好
41%
Official Table 1; think-high; no tools; Pass@1
成绩出处RefSpatialBench厂商自报 · 越高越好
56.3%
Official Table 2; think-high; Pass@1
成绩出处SimpleVQA厂商自报 · 越高越好
65.4%
Official Table 2; think-high; Pass@1
成绩出处SuperGPQA厂商自报 · 越高越好
64.8%
Official Table 1; think-high; no tools; Pass@1
成绩出处SWE-Bench Verified厂商自报 · 越高越好
72.9%
Official Table 4; think-high; benchmark-specific agent tools
成绩出处TempCompass厂商自报 · 越高越好
86.9%
Official Table 3; think-high; public video evaluation
成绩出处Terminal-Bench 2.0厂商自报 · 越高越好
45.2%
Official Table 4; think-high; benchmark-specific agent tools
成绩出处TOMATO厂商自报 · 越高越好
60.8%
Official Table 3; think-high; public video evaluation
成绩出处TVBench厂商自报 · 越高越好
71.5%
Official Table 3; think-high; public video evaluation
成绩出处VideoMME w sub.厂商自报 · 越高越好
87.8%
Official Table 3; think-high; subtitles included
成绩出处VideoMMMU厂商自报 · 越高越好
82.7%
Official Table 3; think-high; public video evaluation
成绩出处VideoSimpleQA厂商自报 · 越高越好
67.8%
Official Table 3; think-high; public video evaluation
成绩出处VLMsAreBiased厂商自报 · 越高越好
62%
Official Table 2; think-high; Pass@1
成绩出处VLMsAreBlind厂商自报 · 越高越好
93%
Official Table 2; think-high; Pass@1
成绩出处WideSearch厂商自报 · 越高越好
63.8%
Official Table 4; think-high; benchmark-specific agent tools
成绩出处ZEROBench厂商自报 · 越高越好
11%
Official Table 2; think-high; Pass@1
成绩出处
