AllSpark's Iris-mini and Iris-pro, built on Qwen3.6-35B-A3B and Qwen3.5-397B-A17B, deliver the strongest results among open-weight search agents in their size classes, per the team's paper. Iris-mini scored 82.2 on BrowseComp; Iris-pro reached 88.6. Training data was reverse-engineered from web link structure, with two-stage filtering and reinforcement learning. Context management boosted Iris-mini's BrowseComp scores by up to 21.2 points. The team reports both models improved on untrained tasks, including general tool use and office work.
No score is assigned. Sources and their independence are shown in the citation chain below.