BrowseComp-Plus
A benchmark for evaluating deep research capabilities of AI agents on complex information retrieval and synthesis tasks. It serves as an extended version of BrowseComp, targeting more demanding scenarios that require multi-step search and comprehensive analytical reasoning.
Core Facts
Timeline (last 90 days)
Under the condition of matched per-step generation budgets, FOCUS at the Qwen3.5-9B scale improves by 7.2 percentage points over the strongest baseline (the higher of GRPO/VinePPO) on the BrowseComp-Plus benchmark, reaching 55.4 versus 48.2.
Drex for Search achieved a score of 92.05% on the BrowseComp-Plus benchmark.
Drex for Search requires an average of only 12.6 search calls on BrowseComp-Plus.
BrowseComp-Plus is a controlled version of BrowseComp that uses a fixed corpus instead of the live web to eliminate evaluation noise caused by network fluctuations and page changes.
On BrowseComp-Plus, attribution-guided SFT matches the effect of full-turn exact-context SFT using only about 30% of the training tokens.
On BrowseComp-Plus, attribution-guided PPO is 4.1 percentage points higher than PPO using only outcome rewards.
On the BrowseComp-Plus benchmark, CLM paired with Qwen3 27B achieves 59.4%, approximately 11% higher than the best summarization solution, with about one-fifth less computation.
All Facts (8)
BrowseComp is an agentic search benchmark proposed by OpenAI that evaluates a model's ability to complete multi-step information retrieval and integration in a real web environment
70%UnverifiedUnder the condition of matched per-step generation budgets, FOCUS at the Qwen3.5-9B scale improves by 7.2 percentage points over the strongest baseline (the higher of GRPO/VinePPO) on the BrowseComp-Plus benchmark, reaching 55.4 versus 48.2.
50%UnverifiedDrex for Search achieved a score of 92.05% on the BrowseComp-Plus benchmark.
50%UnverifiedDrex for Search requires an average of only 12.6 search calls on BrowseComp-Plus.
50%UnverifiedOn BrowseComp-Plus, attribution-guided SFT matches the effect of full-turn exact-context SFT using only about 30% of the training tokens.
50%UnverifiedOn BrowseComp-Plus, attribution-guided PPO is 4.1 percentage points higher than PPO using only outcome rewards.
50%UnverifiedBrowseComp-Plus is a controlled version of BrowseComp that uses a fixed corpus instead of the live web to eliminate evaluation noise caused by network fluctuations and page changes.
50%UnverifiedOn the BrowseComp-Plus benchmark, CLM paired with Qwen3 27B achieves 59.4%, approximately 11% higher than the best summarization solution, with about one-fifth less computation.
50%