AetherCode厂商自报 · 越高越好
29.8%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处AIME 2025厂商自报 · 越高越好
87%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处AIME 2026厂商自报 · 越高越好
86.7%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处ARC-AGI厂商自报 · 越高越好
29.8%
Model Card Table 8; ArcAGI1-Image percentage score.
成绩出处ArcAGI2厂商自报 · 越高越好
2.3%
Model Card Table 4 efficient-models; ARC-AGI-2 text/general-reasoning percentage score.
成绩出处ARC-AGI v2厂商自报 · 越高越好
1.5%
Model Card Table 8; ArcAGI2-Image percentage score.
成绩出处BabyVision厂商自报 · 越高越好
38.7%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处Beyond AIME厂商自报 · 越高越好
69%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处BLINK厂商自报 · 越高越好
73.4%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处ChartQAPro厂商自报 · 越高越好
65.2%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处CharXiv-D厂商自报 · 越高越好
91.9%
Model Card Table 8; CharXiv-DQ percentage score.
成绩出处CharXiv-R厂商自报 · 越高越好
70.8%
Model Card Table 8; CharXiv-RQ percentage score.
成绩出处CL-bench厂商自报 · 越高越好
14.8%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处CodeForces厂商自报 · 越高越好
1,644分
Model Card Table 4 efficient-models; Codeforces Elo rating.
成绩出处COLLIE厂商自报 · 越高越好
91.2%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处ContPhy厂商自报 · 越高越好
55.9%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处CountBench厂商自报 · 越高越好
0.96%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处CrossVid厂商自报 · 越高越好
58.6%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处DUDE厂商自报 · 越高越好
68.9%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处DynaMath厂商自报 · 越高越好
58.9%
Model Card Table 8; DynaMath worst-case accuracy over 10 variants.
成绩出处EMMA厂商自报 · 越高越好
57%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处ERQA厂商自报 · 越高越好
56.3%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处FActScore厂商自报 · 越高越好
50.4%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处FRAMES厂商自报 · 越高越好
80.5%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处FrontierScience Olympiad厂商自报 · 越高越好
44%
Model Card Table 4 efficient-models; FrontierScience olympiad split percentage score.
成绩出处FrontierScience Research厂商自报 · 越高越好
3.3%
Model Card Table 4 efficient-models; FrontierScience research split percentage score.
成绩出处Global PIQA厂商自报 · 越高越好
89.2%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处GPQA厂商自报 · 越高越好
79%
Model Card Table 4 efficient-models; GPQA Diamond percentage score mapped to existing GPQA convention.
成绩出处Graphwalks BFS <128k厂商自报 · 越高越好
64.1%
Model Card Table 4 efficient-models; Graphwalks BFS <128K percentage score.
成绩出处Graphwalks parents <128k厂商自报 · 越高越好
93%
Model Card Table 4 efficient-models; Graphwalks Parents <128K percentage score.
成绩出处Hallusion Bench厂商自报 · 越高越好
65.1%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处HealthBench厂商自报 · 越高越好
30%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处HealthBench Hard厂商自报 · 越高越好
15.3%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处HiPhO厂商自报 · 越高越好
55.8%
Model Card Table 8; HiPhO average normalized score over 13 physics olympiad competitions.
成绩出处Humanity's Last Exam (no tools, text-only)厂商自报 · 越高越好
13.3%
Model Card Table 4 efficient-models; no-tools, text-only HLE percentage score.
成绩出处LiveCodeBench v6厂商自报 · 越高越好
64.1%
Model Card Table 4 efficient-models; LiveCodeBench v6 percentage score.
成绩出处LiveSports-3K厂商自报 · 越高越好
73.3%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处LongFact Concepts厂商自报 · 越高越好
91.4%
Model Card Table 4 efficient-models; LongFact Concepts percentage score.
成绩出处LongFact Objects厂商自报 · 越高越好
87%
Model Card Table 4 efficient-models; LongFact Objects percentage score.
成绩出处LongVideoBench厂商自报 · 越高越好
74.8%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处LVBench厂商自报 · 越高越好
66.6%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处MathArena Apex厂商自报 · 越高越好
4.2%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处MathVision厂商自报 · 越高越好
78.1%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处MathVista-Mini厂商自报 · 越高越好
85.5%
Model Card Table 8; MathVista testmini Pass@1 percentage score.
成绩出处Minerva厂商自报 · 越高越好
54.7%
Model Card Table 9 public video benchmarks; subtitle-marked Minerva score.
成绩出处MMLongBench-Doc厂商自报 · 越高越好
0.49%
Model Card Table 8; MMLongBench-Doc percentage score.
成绩出处MMLU-Pro厂商自报 · 越高越好
83.6%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处MMMLU厂商自报 · 越高越好
81.6%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处MMMU厂商自报 · 越高越好
79.7%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处MMMU-Pro厂商自报 · 越高越好
71.4%
Model Card Table 8; MMMU-Pro aggregate over Standard 10-option and Vision subsets.
成绩出处MMStar厂商自报 · 越高越好
79.1%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处MMVU厂商自报 · 越高越好
69%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处MotionBench厂商自报 · 越高越好
64.4%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处MRCR v2 (8-needle)厂商自报 · 越高越好
51.4%
Model Card Table 4 efficient-models; MRCR v2 8-needle percentage score.
成绩出处MTVQA厂商自报 · 越高越好
50.6%
Model Card Table 8; MTVQA scored by DeepSeek-V3-0324 LLM judge.
成绩出处MuirBench厂商自报 · 越高越好
78%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处Multi-Challenge厂商自报 · 越高越好
61.1%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处OCRBench_V2厂商自报 · 越高越好
58.5%
Model Card Table 8; OCRBenchv2 average of overall English and Chinese scores.
成绩出处OmniDocBench 1.5厂商自报 · 越高越好
11%
Model Card Table 8; OmniDocBench 1.5 overall edit distance.
成绩出处OVBench厂商自报 · 越高越好
60.1%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处OVOBench厂商自报 · 越高越好
70.4%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处PHYBench厂商自报 · 越高越好
56%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处RealWorldQA厂商自报 · 越高越好
81.6%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处RefSpatialBench厂商自报 · 越高越好
0.56%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处SimpleQA Verified厂商自报 · 越高越好
18.9%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处SimpleVQA厂商自报 · 越高越好
0.69%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处SuperChem厂商自报 · 越高越好
16.2%
Model Card Table 4 efficient-models; SuperChem text-only percentage score.
成绩出处SuperGPQA厂商自报 · 越高越好
61.6%
Model Card Table 4 efficient-models; reported percentage score.
成绩出处TempCompass厂商自报 · 越高越好
83.7%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处TOMATO厂商自报 · 越高越好
47.4%
Model Card Table 9 public video benchmarks; default TOMATO score.
成绩出处TreeBench厂商自报 · 越高越好
57.3%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处TVBench厂商自报 · 越高越好
70.5%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处VideoHolmes厂商自报 · 越高越好
58.6%
Model Card Table 9 public video benchmarks; subtitle-marked VideoHolmes score.
成绩出处VideoMME w sub.厂商自报 · 越高越好
81.2%
Model Card Table 9 public video benchmarks; VideoMME with subtitles.
成绩出处VideoMMMU厂商自报 · 越高越好
80.6%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处VideoSimpleQA厂商自报 · 越高越好
67.7%
Model Card Table 9 public video benchmarks; reported percentage score.
成绩出处VisFactor厂商自报 · 越高越好
23.6%
Model Card Table 8; VisFactor macro-average percentage score.
成绩出处VisuLogic厂商自报 · 越高越好
0.4%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处VLMsAreBiased厂商自报 · 越高越好
58.4%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处VLMsAreBlind厂商自报 · 越高越好
93.1%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处WorldVQA厂商自报 · 越高越好
47.6%
Model Card Table 8 public visual-language benchmarks; Pass@1 percentage score.
成绩出处ZEROBench厂商自报 · 越高越好
0.07%
Model Card Table 8; ZEROBench main split percentage score.
成绩出处ZEROBench-Sub厂商自报 · 越高越好
0.36%
Model Card Table 8; ZEROBench sub split percentage score.
成绩出处
