Salesforce trains 8B model to generate diverse research ideas, reports it beats Claude Opus 4.8 and GPT-6 Astra

Salesforce and colleagues trained an 8B reasoning model that writes several research ideas in one generation from a set of source papers.

Its reward scores the whole set, weighting each idea's quality by how far it sits from the ideas written before it, so the model learns to diversify its output. On the team's contamination-free benchmark, the 8B model scores higher on creative utility, a measure combining quality and diversity, than Claude Opus 4.8 and GPT-6 Astra, according to a report by Elvis Saravia on X. The report gives no scores, and the benchmark results come from the researchers' own evaluation.