cruxeval_input_cot: 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)
qwen2.5-coder-32b-instruct 76.1 92.8 44.6 10 1.5 1.2 0.97
google_gemma_3_27b_it 69 87 38.7 9 1.6 1.3 1
qwen2.5-coder-14b-instruct 66.1 91 37 12 1.7 1.2 1.1
google_gemma_3_12b_it 62.5 85.6 33.7 11 1.7 1.3 1.1
llama-3.1-70B-instruct 61.4 61.4 32.9 13 1.7 1.7 0
google_gemma_2_27b_it 58 85 30.6 10 1.7 1.3 1.1
qwen2-72b-instruct 56.9 84.2 29.8 10 1.8 1.3 1.2
qwen2-math-72b-instruct 56.6 84.9 29.8 10 1.8 1.3 1.2
mistralai_mixtral_8x22b_instruct_v0.1 56.1 86.8 29.6 11 1.8 1.2 1.3
qwen2.5-coder-7b-instruct 54.9 85.4 28.8 11 1.8 1.3 1.2
google_gemma_2_9b_it 47.9 77 23.4 12 1.8 1.4 1.1
deepseek_r1_distill_qwen_14b 47 78.2 23.8 11 1.8 1.3 1.2
qwen1.5-72b-chat 47 79.5 23.3 10 1.8 1.3 1.2
deepseek_r1_distill_llama_70b 46.8 79.1 23.6 10 1.8 1.3 1.2
google_codegemma_1.1_7b_it 44.7 82.6 22.2 13 1.8 1.2 1.3
mistralai_mathstral_7b_v0.1 44 78.2 21.2 11 1.8 1.2 1.2
google_gemma_3_4b_it 43.7 79.8 21.6 13 1.8 1.3 1.2
mistralai_ministral_8b_instruct_2410 43.3 77.9 20.8 11 1.8 1.2 1.2
qwen2.5-coder-3b-instruct 43.3 82.9 21.7 12 1.8 1.1 1.3
llama-3.1-8B-instruct 41.9 42 20.5 15 1.7 1.7 0.052
qwen1.5-32b-chat 41.4 77.2 19.8 11 1.7 1.2 1.2
qwen2-7b-instruct 40.4 76.9 19 11 1.7 1.2 1.2
qwen1.5-14b-chat 39.6 73.5 18.7 12 1.7 1.2 1.2
qwen3-1.7b 38.6 72.1 18.4 12 1.7 1.2 1.2
qwen3-4b 37.9 64.8 18.1 12 1.7 1.3 1.1
deepseek_r1_distill_llama_8b 36 71.5 16.7 12 1.7 1.2 1.2
mistralai_mixtral_8x7b_instruct_v0.1 34.5 75.2 16.1 12 1.7 1.1 1.3
qwen2.5-coder-1.5b-instruct 34.3 78 16.9 12 1.7 1 1.3
qwen3-14b 32.3 51.6 14.9 12 1.7 1.4 0.89
llama-3.2-3B-instruct 32.1 32.1 15.4 18 1.7 1.7 0
qwen3-32b 31.9 56.8 14.9 11 1.6 1.3 0.98
mistralai_mistral_7b_instruct_v0.3 31.7 65.6 14 11 1.6 1.2 1.1
qwen2-math-7b-instruct 31.1 65 14.8 12 1.6 1.1 1.2
google_gemma_7b_it 29.7 63.4 14.6 12 1.6 1.1 1.2
qwen2.5-coder-0.5b-instruct 28.2 72.8 14.7 13 1.6 0.91 1.3
qwen1.5-7b-chat 27.2 65.5 12 12 1.6 1.1 1.2
deepseek_v2_lite_chat 26.5 65.6 12 11 1.6 1 1.2
mistralai_mistral_7b_instruct_v0.1 26.3 63.1 12 11 1.6 1 1.2
mistralai_mistral_7b_instruct_v0.2 24.7 60.2 10.7 11 1.5 0.98 1.2
qwen3-0.6b 22.7 59.9 10.6 13 1.5 0.9 1.2
qwen3-8b 21.2 47.5 9.66 12 1.4 0.99 1.1
google_gemma_2b_it 18.7 52.1 9.76 13 1.4 0.85 1.1
deepseek_r1_distill_qwen_7b 16.9 52.5 7.15 11 1.3 0.74 1.1
qwen2-math-1.5b-instruct 16.9 46.4 7.28 12 1.3 0.83 1
qwen2-1.5b-instruct 15.1 60.1 7.04 12 1.3 0.58 1.1
llama-3.2-1B-instruct 8.88 8.88 4.3 21 1 1 0
deepseek_r1_distill_qwen_1.5b 6.12 27.3 2.09 12 0.85 0.42 0.74
qwen2-0.5b-instruct 4.86 29.1 2.18 13 0.76 0.27 0.71
qwen1.5-1.8b-chat 3.06 24.4 1.42 12 0.61 0.15 0.59
qwen1.5-0.5b-chat 1.73 16.6 0.867 13 0.46 0.11 0.45
deepseek_r1_distill_qwen_32b 1.56 10.8 0.564 10 0.44 0.13 0.42
google_gemma_3_1b_it 0.0865 0.5 0.0354 13 0.1 0.055 0.088