ML Engineering on MLE-Bench official (test) benchmark leaderboard · SOTA2 Research

ML Engineering on MLE-Bench official (test)

71.2Medal Rate (Low)Gome GPT-5 (24h)

Evaluation Results

Method
Gome GPT-5 (24h)2026.03	71.2	28.1	26.7	40.4
Gome GPT-5 (12h)2026.03	68.2	21.1	22.2	35.1
Thesis GPT-5-Codex2026.03	65.2	45.6	31.1	48.4
MLE-STAR Gemini-2.5-pro2026.03	63.6	—	—	—
Gome o3 (24h)2026.03	63.6	27.2	24.4	36.4
InternAgent-MLE DeepSeek-R12026.03	62.1	26.3	24.4	36.4
FM Agent Gemini-2.5-pro2026.03	62.1	36.8	33.3	43.6
Gome o3 (12h)2026.03	59.1	20.2	22.2	32.5
AIRA o32026.03	55	22	21.7	31.6
ML-Master GPT-52026.03	51.5	10.5	17.8	24
KompeteAI Gemini-2.5-flash2026.03	51.5	—	—	—
Gome DeepSeek-R12026.03	50	9.7	20	23.4
Neo Multi Frontier-Tier Models2026.03	48.5	29.8	24.4	34.2
ML-Master DeepSeek-R12026.03	47	9.7	20	22.7
ML-Master o32026.03	42.4	12.3	20	22.7
AIDE o1-preview2026.03	34.3	8.8	10	16.9
AIDE gpt-4o-24-082026.03	19	3.2	5.6	8.6
OpenHands gpt-4o-24-082026.03	11.5	2.2	1.9	5.1
MLAB gpt-4o-24-082026.03	4.2	0	0	1.3