By freely browsing the internet, analyzing responses, and producing thorough documents on any subject, OpenAI’s novel Deep Research agent hopes to revolutionize how users interact with online data.
The AI company demonstrated its capabilities by tackling whatever from alpine purchase recommendations to superior science papers.
But it’s not for the weak. OpenAI restricts access to Pro users who pay$ 200 per month for the privilege.
There’s a reason behind the great rate:” It is very compute-intensive and slow, but it’s the second AI system that can do for a wide variety of complex, significant things”, Sam Altman tweeted.
it is very compute-intensive and slower, but it’s the first ai program that can do for a wide variety of sophisticated, important things.
going lived in our pro tier then, with 100 inquiries per month.
plus, group, and enterprise may come soon, and therefore free tier.
— Sam Altman ( @sama ) February 3, 2025
Yeah, also, that’s not entirely correct. There have been a couple options available for decades.
Users have been able to use AI for challenging study for a while, from open-source initiatives like the AI Scientist to closed-source tasks like Gemini’s Deep Research.
Those who are unwilling to pay the price and those who have a little imagination you produce comparable or better outcomes for less money or for nothing.
Here are some of our major takes for the budget-conscious:
Google Gemini: The 90 % Cheap Option
Google didn’t waited for OpenAI to render its walk. Deep Research was introduced to Gemini Advanced users for$ 20 a month in December by the tech giant.
For that price, Gemini users get access to Gemini Ultra, Gemini’s Deep Research Agent, 2TB of cloud storage, and the ability to build personal agents or “gems” —similar to OpenAI’s GPTs.
In an established statement, the company stated that” Deep Research uses AI to explore difficult matters on your behalf and provide you with results in a thorough, easy-to-read report.” This is a first look at how Gemini is improving its ability to tackle challenging tasks without wasting your time.
Unlike OpenAI’s limiting strategy, Google threw aside comment limits entirely. Gemini people receive endless research calls, compared to OpenAI’s plan to limit Plus people at 10 regular concerns and Pro users at 100.
That means Google’s resource offers 10 times more price at 10 % of the cost.
tested the application and the analysis process mirrored OpenAI’s method. Users enter their subject and receive a analysis plan, just like ChatGPT users. Gemini users may modify the strategy before unleashing the AI on the web, in contrast to OpenAI’s technique, which asks the audience a few questions prior to the research.
When it comes to solutions, Gemini played it safe. Our tests revealed that the model regularly preferred well-known scientific publications and reliable news outlets over smaller blogs or websites.
According to its schedule, the agent comb through dozens of websites at once, which is how it typically takes a few minutes to complete the analysis.
For instance, we requested that the design conduct some research on the Nvidia Jetson Nano. In 3 days it scraped through 40 sites and provided a 6-page statement on the technology, how it worked, its possible, some of the most significant competitors, and various pieces of information
Google’s use over OpenAI has a significant advantage because Gemini is so fully integrated into Google’s ecosystem. Users can easily expand their research using Google Docs by exporting it right to Google Docs. ( There is no currently integration with other Google suite tools like Sheets. )
Overall, Google has probably two of the biggest hidden benefits that this research agent and NotebookLM have.
You.com: A Research Agent For Those Who Favor Speed Over Depth
You.com may not be the most well-known AI site in the world, but it is not to be underestimated.
The four-year-old startup, backed by an impressive roster of tech investors, built its reputation as a Perplexity competitor in the AI-powered search engine space.
It’s not quite a unicorn—yet. You.com’s parent company, SuSea Inc., raised$ 50 million in its latest funding round, pushing its valuation from$ 700 million to$ 900 million.
In November 2024, You.com was at the top of the list of AI companies that offered research agents.
” A shift in how AI conducts research is represented by enabling AI to analyze up to 200 sources per query and employ PhD-level research methodologies,” the company said in an official announcement. We imagined the AI as a diligent student taking a PhD-level course and doing its homework before writing a master’s thesis.
Its research assistant took a stripped-down approach. Instead of OpenAI’s methodical planning phase or Google’s tweakable research outline, You.com cut straight to the chase.
Users typed their query, and the model delivered results, sometimes in mere seconds but never longer than a few minutes. By comparison, the results shown in OpenAI’s demo indicate that its research agent took several minutes to respond, with one instance taking nearly 25 minutes to provide a briefing
This speed-first approach comes with trade-offs. The average length of a You.com report was two to three pages, which is shorter than its rivals but strong enough to support further research. Before requiring users to switch to a different model, the platform offered three free daily uses.
You.com is an ideal starting point for research projects because of its simple process. Although it may not be as comprehensive as Google or OpenAI’s offerings, it shows that expensive subscriptions aren’t always required for effective AI research assistance.
You.com is the only large platform that provides users with a free AI research agent.
Deepseek R1: The Free DIY Powerhouse
Yes, this one is from China, and it’s not exactly an Agent—but it may be a nice workaround for users willing to put in extra effort.
OpenAI’s flagship reasoning model lacks web browsing capabilities. Deepseek’s R1 filled that gap as the only reasoning model with built-in internet access.
We can use this resource for free to produce thorough, in-depth research.
More user input is required, but the setup produces surprisingly accurate results.
Start by turning on Deep Reasoning to start R1 by simply clicking the appropriate button in the text box. This is a necessary step since the default Deepseek V3 concentrates more on creative tasks, similar to GPT-4o, rather than OpenAI o1.
With web search enabled and the reasoning activated, all that is required is to provide the model with detailed instructions on what to look for, ideally organized in the form of a research plan or a set of guidelines.
We tried Deepseek for research and it was very transparent. R1 displayed its work and thought processes in real-time.
It was possible to watch it breaking down queries, outlining approaches, and explaining why certain information mattered. This glimpse into the model’s digital brain could aid in model-based instruction refinement and understanding.
The initial output matched You.com’s scope— a concise report with solid sourcing.
But R1’s real power showed up in follow-up queries. Drill down into specific sections, request more technical details or beginner-friendly explanations and, in general, guide the model toward the research.
Instead of waiting half an hour for a report and having it do everything again, you can iterate with the model and ask it to elaborate on specific topics consistently.
In our test, an initial 3-pager was expanded to over 25 pages of factual, accurate, and relevant content, all within the scope of the topic we researched.
While it required more hands-on guidance than point-and-click alternatives, Deepseek R1 proved that effective AI research assistance doesn’t need a monthly subscription.
Aside from these significant players, there are a few other tools that have created niches in the AI research market. By focusing on follow-up questions, which helped users uncover unanticipated angles in their research topics, Beago AI and Perplexity Pro gained traction. You might want to check out Johns Hopkins University’s and AMD’s collaboration, Agent Laboratory, an open-source solution.
edited by Sebastian Sinclair and Josh Quittner
Generally Intelligent Newsletter
A generative AI model’s generative AI model, Gen, tells a weekly AI journey.