Adaptive Instructed-Retriever: Frontier-Quality Search at 2x Lower Latency

Summary

Adaptive Instructed-Retriever addresses the tension between retrieval quality and latency for enterprise data agents, especially on multi-hop questions that benefit from sequential search. It combines parallel single-step retrieval with bounded multi-step search, using a fixed upper limit and an adaptive stopping policy that spends extra steps only when they are likely to improve results. The model is trained with online reinforcement learning using CISPO, a reward that balances trajectory quality against search cost, and synthetic multi-hop data alongside Instructed-Retriever-1 data. Across seven held-out internal and external benchmarks, it matches leading third-party and open-source models while answering in 5.8 seconds, reported as more than twice as fast as the compared alternatives, and substantially improves over single-step search. Varying the step penalty produces checkpoints spanning different quality-latency tradeoffs for production workloads.

Takeaways

1. 1

   The model can stop early when evidence is sufficient and use additional sequential search only for requests where further retrieval is likely to help.

2. 2

   Training reuses Instructed-Retriever-1 data, adds synthetic multi-hop questions, and uses CISPO to penalize search steps that do not produce corresponding performance gains.

3. 3

   Across seven held-out internal and external benchmarks, the model answers in 5.8 seconds while matching leading compared models and improving substantially over single-step search.