Agents' Last Exam厂商自报 · 越高越好
50.3%
Agents' Last Exam. Max reasoning effort.
成绩出处ARC-AGI-3厂商自报 · 越高越好
0.18%
ARC-AGI-3. Max reasoning effort.
成绩出处Artificial Analysis来源汇总 · 越高越好
51%
Artificial Analysis Intelligence Index v4.1 (max). Score 51.
成绩出处AutomationBench厂商自报 · 越高越好
14.9%
AutomationBench. Max reasoning effort.
成绩出处BenchCAD厂商自报 · 越高越好
63.1%
BenchCAD (no tools). Max reasoning effort.
成绩出处BenchCAD (with Python tool)厂商自报 · 越高越好
73.9%
BenchCAD with Python tool. Max reasoning effort.
成绩出处Big Finance Bench厂商自报 · 越高越好
36%
Big Finance Bench. Max reasoning effort.
成绩出处BrowseComp厂商自报 · 越高越好
83.3%
BrowseComp. Max reasoning effort.
成绩出处Capture-the-Flag Challenges (Internal)厂商自报 · 越高越好
85.2%
Capture-the-Flag Challenges (Internal). Max reasoning effort.
成绩出处Connectors厂商自报 · 越高越好
99.9%
Connectors production benchmark pass rate.
成绩出处DeepSWE厂商自报 · 越高越好
67.2%
DeepSWE v1.1. Max reasoning effort.
成绩出处DeepSWE 1.1来源汇总 · 越高越好
67%
DeepSWE v1.1 leaderboard, mini-swe-agent harness, max effort; Pass@1 67% ± 4%.
成绩出处ExploitBench厂商自报 · 越高越好
33.2%
ExploitBench. Max reasoning effort.
成绩出处ExploitGym厂商自报 · 越高越好
12.4%
ExploitGym, six-hour cap. Max reasoning effort.
成绩出处FrontierCode 1.1来源汇总 · 越高越好
39.8%
FrontierCode 1.1 current leaderboard; mergeability score at max effort: 39.8%.
成绩出处FrontierMath厂商自报 · 越高越好
78.6%
FrontierMath Tier 1-3 (v2). Max reasoning effort.
成绩出处FrontierMath Tier 4 (v2)厂商自报 · 越高越好
58.5%
FrontierMath Tier 4 (v2). Max reasoning effort.
成绩出处GDP.pdf厂商自报 · 越高越好
22.7%
gdp.pdf. Max reasoning effort.
成绩出处GeneBench-Pro厂商自报 · 越高越好
10.8%
GeneBench Pro. Max reasoning effort.
成绩出处GPQA厂商自报 · 越高越好
92.3%
GPQA Diamond. Max reasoning effort.
成绩出处Graphwalks BFS >128k厂商自报 · 越高越好
81.3%
GraphWalks BFS 256k f1. Max reasoning effort.
成绩出处Graphwalks BFS 1M厂商自报 · 越高越好
51.2%
GraphWalks BFS 1M f1. Max reasoning effort.
成绩出处HealthBench厂商自报 · 越高越好
55.8%
HealthBench, length-adjusted score (unadjusted 55.4).
成绩出处HealthBench Consensus厂商自报 · 越高越好
95.1%
HealthBench Consensus, length-adjusted score (unadjusted 95.1).
成绩出处HealthBench Hard厂商自报 · 越高越好
32%
HealthBench Hard, length-adjusted score (unadjusted 31.4).
成绩出处HealthBench Professional厂商自报 · 越高越好
55.7%
HealthBench Professional, length-adjusted score (unadjusted 59.8).
成绩出处Internal Research Debugging Evaluation厂商自报 · 越高越好
50.8%
Internal Research Debugging Evaluation. Max reasoning effort.
成绩出处KernelGen 1P厂商自报 · 越高越好
22.4%
KernelGen 1P. Max reasoning effort.
成绩出处LifeSciBench厂商自报 · 越高越好
51.2%
LifeSciBench. Max reasoning effort.
成绩出处Management Consulting Tasks (Internal)厂商自报 · 越高越好
35.4%
Management Consulting Tasks (Internal). Max reasoning effort.
成绩出处MedChemBench (Internal)厂商自报 · 越高越好
30.4%
MedChemBench (Internal). Max reasoning effort.
成绩出处MMMU-Pro厂商自报 · 越高越好
78.4%
MMMU Pro (no tools). Max reasoning effort.
成绩出处MMMU-Pro (with tools)厂商自报 · 越高越好
79.5%
MMMU Pro (with tools). Max reasoning effort.
成绩出处MRCR v2 (8-needle)厂商自报 · 越高越好
41.3%
OpenAI MRCR v2 8-needle, 256K-512K. Max reasoning effort.
成绩出处MRCR v2 (8-needle, 512K-1M)厂商自报 · 越高越好
41.3%
OpenAI MRCR v2 8-needle, 512K-1M. Max reasoning effort.
成绩出处NanoGPT厂商自报 · 越高越好
1.66%
NanoGPT. Max reasoning effort.
成绩出处OSWorld 2.0厂商自报 · 越高越好
45.6%
OSWorld 2.0 binary completion. Max reasoning effort.
成绩出处PostTrainBench Lite厂商自报 · 越高越好
29.6%
PostTrainBench Lite mean reward. Max reasoning effort.
成绩出处RSI Index厂商自报 · 越高越好
41.9%
RSI Index (aggregate self-improvement). Max reasoning effort.
成绩出处Search and Function-Calling厂商自报 · 越高越好
89.7%
Search and Function-Calling production benchmark pass rate.
成绩出处SEC-bench Pro厂商自报 · 越高越好
48.9%
SEC-Bench Pro. Max reasoning effort.
成绩出处SWE-Bench Pro厂商自报 · 越高越好
62.7%
SWE-Bench Pro. Max reasoning effort.
成绩出处Terminal-Bench 2.1厂商自报 · 越高越好
84.7%
Terminal-Bench 2.1. Max reasoning effort.
成绩出处Terminal-Bench 4.0来源汇总 · 越高越好
17.3%
Terminal-Bench 4.0 public leaderboard; resolution rate 17.3% ± 2.8%; Codex agent; max effort.
成绩出处Toolathlon厂商自报 · 越高越好
53.4%
Toolathlon. Max reasoning effort.
成绩出处