List your company

Scientific Ideation on Scientific Ideation 60 Samples Human-Validated (test)

3.63NoveltyGround Truth
2.15322.53662.923.3034Mar 19, 2026

Evaluation results

MethodLinks
Ground Truth2026.033.633.693.443.58—
Claude-3.5-SonnetBackbone=claude-3-5-sonnet-2024-10-222026.033.393.072.823.093.2
MoRIws=0.7, we=0.3, entropy mask=Top-25%, training steps=4002026.033.313.163.113.19—
Full-SFT2026.033.12.873.022.996.7
AI-Scientist-V2(Claude)Backbone=claude-3-5-haiku-2024-10-222026.032.742.462.892.718.1
ResearchAgentBackbone=gpt-4o-2024-08-062026.032.662.512.632.622.7
GPT-4oBackbone=gpt-4o-2024-08-062026.032.512.782.792.6918.6
AI-Scientist-V2(GPT)Backbone=gpt-4o-2024-08-062026.032.482.672.982.7117.7
VirSciBackbone=gpt-4o-2024-08-062026.032.212.262.282.2541.8