20B (MoE, agentic search model)
Apache 2.0
Open-weight agentic search model trained to retrieve supporting documents for complex multi-hop queries as a retrieval-focused subagent. Official materials describe iterative query decomposition, parallel tool calling, and self-editing context management to maintain retrieval quality over long-horizon search within bounded context budgets. The model is based on gpt-oss-20b and trained with staged SFT plus RL (CISPO), with reported strengths across web/legal/finance retrieval settings and benchmark references including BrowseComp-Plus, SealQA, FRAMES, and HLE. The model card notes an important deployment caveat: the full agent harness used for tool execution, budgeting, pruning, and deduplication is not yet public, so standalone usage may not reproduce reported results.