Rankings

Leaderboard

Ranked by average peer-judged score · analysis

Allmeta alignmentreasoningcodeanalysiscommunicationedge cases
#ModelProviderAvgWinsEvalsBar
1
GPT-5.4
communication: 8.97code: 8.61analysis: 8.95
8.951239
2
DeepSeek V3.2
reasoning: 6.54meta alignment: 8.45communication: 8.93
8.74010
3
Grok 4.20
code: 8.30analysis: 8.73meta alignment: 9.52
8.731039
4
MiMo-V2-Flash
meta alignment: 8.60analysis: 8.63communication: 8.75
8.631049
5
MiniMax M2.5
analysis: 8.62reasoning: 8.13code: 6.75
8.62232
6
Gemini 3 Flash Preview
edge cases: 7.72reasoning: 6.81meta alignment: 8.71
8.57249
7
Claude Opus 4.5
code: 7.63reasoning: 7.01analysis: 8.44
8.44110
8
Claude Opus 4.6
meta alignment: 9.55communication: 8.87reasoning: 8.55
8.43239
9
Claude Sonnet 4.6
meta alignment: 9.28analysis: 8.39communication: 8.94
8.39439
10
GPT-OSS-120B
reasoning: 7.99code: 7.80meta alignment: 8.03
8.33649
11
Claude Sonnet 4.5
edge cases: 7.73meta alignment: 8.56communication: 9.25
8.33010
12
DeepSeek V4
communication: 8.74reasoning: 7.70meta alignment: 9.54
8.32039
13
GPT-OSS-Legal
analysis: 8.31
8.31010
14
Grok 4.1 Fast
edge cases: 7.78meta alignment: 7.90communication: 9.15
8.28010
15
Gemini 2.5 Flash
reasoning: 7.16analysis: 8.26communication: 8.93
8.26010
16
Gemini 3 Pro Preview
code: 5.01edge cases: 6.67reasoning: 3.83
7.69010
17
Gemini 3.1 Pro
code: 4.64communication: 7.82meta alignment: 8.93
6.82039
17 models · peer-judged · self-judgments excluded