Apple's SSR framework cuts multimodal search agent reasoning latency by over 90%

Apple researchers introduce Selection-based Structured Reasoning (SSR), which has multimodal search agents select from pre-specified reasoning candidates instead of generating freeform reasoning.

SSR reduces per-turn reasoning latency by over 90% and total per-question inference latency by 28–54% on 2B and 4B models.

The authors say its average success rate is competitive with leading search agents of the same scale across seven multimodal search benchmarks.