The Optimizer Is the Agent: Reasoning-Driven Search across Prompts, Programs, and ML Workflows

Open paper

Benchmark	Model	Rank	Results
abstract-reasoning-on-arc-agi-2	ReASearch	#12	Accuracy: 50.0
arithmetic-reasoning-on-gsm8k	ReASearch (Llama3.1-8B optimized prompt)	#130	Accuracy: 83.4