Title/abstract screening

?Effort:medium

Model	?Reward	Recall	Precision
Claude Opus 5	0.832	0.964	0.591
GPT-5.6 Sol	0.779	0.898	0.599
Claude Sonnet 5	0.752	0.891	0.565
GPT-5.6 Terra	0.730	0.837	0.667
GPT-5.6 Luna	0.640	0.774	0.602

Full-text extraction

?Effort:medium

Model	?F1
Claude Opus 5	0.415
Claude Sonnet 5	0.395
GPT-5.6 Terra	0.372
GPT-5.6 Sol	0.347
GPT-5.6 Luna	0.264

Risk of Bias

?Effort:medium

Model	?Cohen’s κ
GPT-5.6 Luna	0.293
GPT-5.6 Sol	0.252
Claude Sonnet 5	0.210
GPT-5.6 Terra	0.209
Claude Opus 5	0.178