lcb_codegen_v6_080124: by models

Home Paper Code


SE predicted by accuracy

The typical standard errors between pairs of models on this dataset as a function of the absolute accuracy.

SE vs accuracy difference

The standard error of each model pair against their observed accuracy difference. Pairs below a reference line differ by more than the corresponding number of standard errors, i.e. they are statistically distinguishable at that level.

CDF of question level accuracy

Results table by model

model pass1 pass@count win_rate count SE(A) SE_x(A) SE_pred(A)
O4-Mini (High) 80.2 80.2 27.7 1 1.9 NaN NaN
O3 (High) 75.8 75.8 24.5 1 2 NaN NaN
O4-Mini (Medium) 74.2 74.2 22.4 1 2.1 NaN NaN
Gemini-2.5-Pro-06-05 73.6 73.6 21.8 1 2.1 NaN NaN
DeepSeek-R1-0528 73.1 87.9 21.7 10 2.1 1.8 1.1
Gemini-2.5-Pro-05-06 71.8 71.8 20.3 1 2.1 NaN NaN
EXAONE-4.0-32B 70 80.6 19.2 4 2.1 1.8 1.2
OpenReasoning-Nemotron-32B 69.8 83 19.2 10 2.2 1.8 1.2
Gemini-2.5-Pro-03-25 67.8 67.8 18.2 1 2.2 NaN NaN
O3-Mini-2025-01-31 (High) 67.4 67.4 18.3 1 2.2 NaN NaN
Grok-3-Mini (High) 66.7 66.7 21.3 1 2.2 NaN NaN
Qwen3-235B-A22B 65.9 65.9 16.5 1 2.2 NaN NaN
O4-Mini (Low) 65.9 65.9 16.4 1 2.2 NaN NaN
XBai-o4-medium 65 65 15.9 1 2.2 NaN NaN
O3-Mini-2025-01-31 (Med) 63 63 15.2 1 2.3 NaN NaN
Gemini-2.5-Flash-05-20 61.9 61.9 14.1 1 2.3 NaN NaN
Gemini-2.5-Flash-04-17 60.6 60.6 13.4 1 2.3 NaN NaN
O3-Mini-2025-01-31 (Low) 57 57 11.5 1 2.3 NaN NaN
Claude-Opus-4 (Thinking) 56.6 56.6 10.8 1 2.3 NaN NaN
Claude-Sonnet-4 (Thinking) 55.9 55.9 10.8 1 2.3 NaN NaN
QwQ-32B_temp 55.7 55.7 10.8 1 2.3 NaN NaN
Claude-3.7-Sonnet 50.4 50.4 8.48 1 2.3 NaN NaN
Gemini-Flash-2.0-Thinking-01-21 48.9 48.9 8.54 1 2.3 NaN NaN
Gemini-Flash-2.0-Thinking-12-19 48.7 48.7 8.84 1 2.3 NaN NaN
Claude-Sonnet-4 47.1 47.1 7.24 1 2.3 NaN NaN
Claude-Opus-4 46.9 46.9 7.19 1 2.3 NaN NaN
Claude-3.5-Sonnet-20241022 36.4 41.6 3.84 10 2.3 2.2 0.68
Gemini-Flash-2.0-Exp 31.3 31.3 2.61 1 2.2 NaN NaN
GPT-4O-2024-08-06 29.5 38.8 2.02 10 2.1 2 0.81
GPT-4-Turbo-2024-04-09 28.7 37.9 1.91 10 2.1 2 0.82
GPT-4O-mini-2024-07-18 27.5 34.4 1.72 10 2.1 2 0.62
DeepSeek-V3 27.2 35.7 2.83 10 2.1 2 0.75
Claude-3-Haiku 20.2 24 1.04 10 1.9 1.8 0.54