The Jina AI CEO Han Xiao has replicated OpenAI’s newly launched Deep Research system within hours of its debut, releasing an open-source alternative that demonstrates the accelerating pace of AI innovation. The project, node-DeepResearch, uses a modular architecture to mirror OpenAI’s agentic search-and-analysis capabilities at minimal cost. For context, Jina AI is a neural search technology firm. While initial repository setup began eight days ago, GitHub logs show Xiao began significant work on the implementation shortly after Deep Research’s launch on February 2nd, with major features merged by early morning February 3rd.
“OpenAI’s Deep Research is just a search+read+reasoning in a while-loop, right? unless i’m missing miss something, here is my replicate of it in nodejs, using gemini-flash and jina reader,” he wrote on X, the site formerly known as Twitter.
Xiao’s implementation combines Gemini Flash (Google’s lightweight LLM) for reasoning, Jina Reader for webpage parsing, and Brave/DuckDuckGo for search. The system iteratively searches, reads, and synthesizes answers through a while-loop process, handling complex queries like “Who is bigger: Cohere, Jina AI, or Voyage?” in 13 steps. GitHub logs show Xiao initiated the project shortly after Deep Research’s launch on February 2, with major features merged through the early morning hours of February 3.
Feature | OpenAI Deep Research | Jina AI’s clone |
---|---|---|
Cost | $200/month (currently Pro plan required) | Open-source (API key costs) |
Core Tech | Proprietary fine-tuned model | Off-the-shelf APIs |
Agentic Runtime | 5–30 minutes autonomous runs | Configurable token budgets |
node-DeepResearch quickly gained traction in the developer community, accumulating 514 stars and 52 forks within its first hours on GitHub. The repository’s development was primarily thanks to Han Xiao himself, with automated support from a single integration bot.
OpenAI’s Deep Research currently costs $200/month as part of the ChatGPT Pro plan. Its use is limited to 100 queries per month, with enterprise plans coming later.
A side-by-side performance comparison comparing the performance of the two applications was unavailable at the time of writing.
Tell Us What You Think!
You must be logged in to post a comment.