AA-LCR厂商自报 · 越高越好
80%
High reasoning.
成绩出处AIME 2026厂商自报 · 越高越好
94.7%
High reasoning.
成绩出处Beam 128K厂商自报 · 越高越好
65.1%
High reasoning; 128K context.
成绩出处CharXiv-R厂商自报 · 越高越好
78.8%
High reasoning; CharXiv reasoning split.
成绩出处CI Memories Coverage厂商自报 · 越高越好
64.8%
High reasoning; coverage.
成绩出处CI Memories Violation Rate厂商自报 · 越高越好
26.4%
High reasoning; violation rate (lower is better).
成绩出处DeepSearchQA厂商自报 · 越高越好
74.6%
High reasoning.
成绩出处GAIA2厂商自报 · 越高越好
43.3%
High reasoning.
成绩出处GPQA厂商自报 · 越高越好
83.5%
High reasoning; Diamond split; AA evaluation.
成绩出处Humanity's Last Exam厂商自报 · 越高越好
22%
High reasoning; text only; no tools.
成绩出处IFBench厂商自报 · 越高越好
77%
High reasoning.
成绩出处MCP Atlas厂商自报 · 越高越好
75.5%
High reasoning; public set.
成绩出处MMMU-Pro厂商自报 · 越高越好
74%
High reasoning.
成绩出处OmniDocBench 1.5厂商自报 · 越高越好
75.8%
High reasoning.
成绩出处OSWorld-Verified厂商自报 · 越高越好
65.9%
High reasoning.
成绩出处SciCode厂商自报 · 越高越好
43.6%
High reasoning.
成绩出处ScreenSpot Pro厂商自报 · 越高越好
75.4%
High reasoning.
成绩出处Siren AgentDojo Attack Success Rate厂商自报 · 越高越好
28.4%
High reasoning; attack success rate (lower is better).
成绩出处Siren AgentDojo Utility厂商自报 · 越高越好
94.2%
High reasoning; utility.
成绩出处SkillsBench厂商自报 · 越高越好
44.3%
High reasoning; with skills.
成绩出处SWE-Bench Pro厂商自报 · 越高越好
51.2%
High reasoning.
成绩出处SWE-Bench Verified厂商自报 · 越高越好
76%
High reasoning.
成绩出处Tau3 Banking厂商自报 · 越高越好
23.5%
High reasoning.
成绩出处Terminal-Bench 2.1厂商自报 · 越高越好
51.7%
High reasoning; Terminus-2 harness.
成绩出处WildClawBench厂商自报 · 越高越好
47.6%
High reasoning.
成绩出处