Judge Agreement and Model Ranking on LMArena 140K benchmark leaderboard · SOTA2 Research

Judge Agreement and Model Ranking on LMArena 140K

0.213Kappa (κ)Gemma4-26B-A4B

Evaluation Results

Method	Links
Gemma4-26B-A4B2026.06		0.213	0.6	77.5	19.2	0.938	0.945	75
Qwen3.5-27B2026.06		0.187	0.51	69.3	20	0.938	0.923	71
Gemma4-E4B2026.06		0.177	0.41	88	18.8	0.924	0.921	79
DeepSeek-V3.22026.06		0.173	0.4	70.9	17.4	0.929	0.93	75
GPT-OSS-120B2026.06		0.156	0.56	53.7	24	0.9	0.901	55
Llama-3.3-70B2026.06		0.131	0.47	56.7	31.1	0.754	0.752	45
GPT-OSS-20B2026.06		0.058	0.51	67.5	22.8	0.9	0.908	66