Tongyi DeepResearch: fully open-source Web Agent matching OpenAI performance with 30B (Activated 3B) parameters. Scores: 32.9, 45.3, 75.0 on key benchmarks.
1/7 We're launching Tongyi DeepResearch, the first fully open-source Web Agent to achieve performance on par with OpenAI's Deep Research with only 30B (Activated 3B) parameters! Tongyi DeepResearch agent demonstrates state-of-the-art results, scoring 32.9 on Humanity's Last Exam, 45.3 on BrowseComp, and 75.0 on the xbench-DeepSearch benchmark.
Real-time analysis of public opinion and engagement
What the community is saying — both sides
Open-source triumph — Many readers are ecstatic that a 30B model can match proprietary systems, praising the project as a game-changer for accessibility and customization and celebrating the release with excitement and thanks.
people want details on how the agent manages multi-step workflows, prevents context overload, and ensures dependable tool calls.
cost against OpenAI, Claude, and Google.
g. , Gaode Mate, legal research), with readers asking where adoption will take off first and suggesting AR, navigation, and enterprise legal use-cases.
Community validation — Many propose community-driven challenges, benchmarks, and collaborative testing on the GitHub repo and Hugging Face to push capabilities and vet reliability.
Recognition of impact — Commenters emphasize how a performant open-source web agent lowers barriers and could accelerate innovation across builders and labs.
Next-step curiosity — Folks are eager to try the model, ask about roadmaps, pricing, and demos, and some offer to help run or host tests to stress real-world behavior.
Several replies cheer the Qwen team for shipping weekly—short, enthusiastic reactions (e.g., "🔥") emphasize momentum and appreciation for fast iteration.
Comments such as "China once again cooks USA" and "US so cooked" frame the discussion as a tech rivalry, celebrating perceived Chinese dominance.
One-liners like "ngmi or wagmi?", "wen moon?", and "I got distracted after the first sentence" inject crypto-style shorthand, jokes, and mild indifference into the thread.
A reply cautions that constantly comparing Chinese models to OpenAI “isn’t a good look” and expresses doubt they’ll surpass OpenAI, urging creators not to cling to another company for attention.
Most popular replies, ranked by engagement
7/7🚀 Try it out! Tongyi DeepResearch agent marks a significant step towards AI that can autonomously turn information into insight. We're open-sourcing the model, framework, and complete solutions to empower the community. Dive in and build with us! 🔗 Homepage:
5/7📈 Performance (Heavy Mode). For maximum capability, our "Heavy Mode" uses the Research-Synthesis framework. Multiple agents research in parallel using our IterResearch paradigm, preventing context overload. A final agent synthesizes their findings, pushing our 30B (A3B)
6/7 🌐 Application. Tongyi DeepResearch agent is already powering real-world applications. 🔹 Gaode Mate: An AI copilot in the Amap navigation app for planning complex multi-day trips. 🔹 Tongyi FaRui: A legal research agent that analyzes case law and statutes, providing
China once again cooks USA
US so cooked
So qwen team is shipping every week 🔥