DeepResearch

Why we included this project

Tongyi DeepResearch is an agentic model that runs the whole research loop on its own: it plans search queries, reads and summarizes web pages, and iterates over several rounds before producing a final answer. So instead of a single-shot chatbot reply, you get a multi-step report assembled from sources the agent actually visited. That suits analysts, researchers, and product teams who want autonomous information gathering. The repository ships the model weights, an inference harness, and evaluation scripts, and it offers both a lightweight ReAct mode and a heavier test-time-scaling mode for harder questions. Because it is Apache-2.0 licensed and runs locally, you can wire it into your own pipelines and swap in your preferred search and page-reading providers. The included benchmarks and data formats give you a way to compare behavior before you commit to a deployment.