GPT Researcher
GPT Researcher is a free open source AI research agent that cites 20+ web sources per run, with deep research mode and support for Llama 4 and Qwen 3.6.
About GPT Researcher
What is GPT Researcher (2026 update)
GPT Researcher is an open source autonomous AI research agent that plans a research task, searches and validates dozens of web sources in parallel, and returns a cited report instead of a single unverified answer.
What is new in 2026
GPT Researcher remains entirely open source under the MIT license with no hosted SaaS version, and the project has kept pace with newer open weight models, adding support for Llama 4 Maverick and Scout as well as Qwen 3.6, both of which have native tool-calling and Qwen 3.6’s context window reaches up to 262,000 tokens. Setting report_type to deep now activates a recursive tree-based exploration mode with configurable breadth, depth, and concurrency for more thorough multi-step research, and report_source can be set to hybrid to combine live web search with a user’s own local documents in the same report. GPT Researcher was also ranked first on Carnegie Mellon’s DeepResearchGym benchmark in May 2025, a result the project continues to cite as evidence of its research quality going into 2026.
Key features
- Autonomous research agent that scrapes, filters, and aggregates 20+ web sources per task
- Deep research mode with recursive, tree-based exploration and configurable depth
- Hybrid mode that blends live web search with local documents in one report
- Support for multiple LLM providers, including OpenAI, Anthropic, and Google, plus retrievers like Tavily and Bing
- Output as plain text, Markdown, PDF, DOCX, or JSON with inline citations
- Streaming progress over WebSocket or Server-Sent Events, plus an official MCP server
Pricing in 2026
GPT Researcher itself has no license fee or subscription. It is self-hosted software, so the only ongoing costs are the LLM and search/retriever providers you connect to it, such as OpenAI or Tavily.
| Plan | Price | What you get |
|---|---|---|
| Open source (self-hosted) | Free | Full source code under MIT license, run locally, as a FastAPI server, or via MCP |
| Underlying LLM usage | Pay-as-you-go, check the official site of your chosen provider | Cost of the language model that powers each research run, roughly $0.01 per run on GPT-4-class usage |
| Underlying retriever usage | Pay-as-you-go, check the official site of your chosen provider | Cost of the web search/retriever service, such as Tavily or Bing, used to gather sources |
Who should use it
GPT Researcher fits developers and technical teams who want to build a research or reporting feature into their own product without paying for a proprietary research API, since the entire codebase is free to inspect, modify, and self-host. It also suits analysts, students, and researchers comfortable running Python who want a cited, multi-source report instead of a single chatbot answer, and teams that want to mix their own internal documents with live web research using the hybrid mode.
Limitations and alternatives
Because there is no hosted version, using GPT Researcher requires setting up API keys for an LLM provider and a retriever, and some technical comfort with Python or running a FastAPI server. Costs are variable rather than fixed, so heavy deep-research usage across many LLM calls can add up depending on the model chosen, check the official site of each provider for current rates. For alternatives, compare GPT Researcher against Perplexity for a polished consumer-facing research assistant, and against Stanford STORM or Local Deep Research if a different open source research architecture better fits your stack.
Frequently asked questions
Is GPT Researcher free?
Yes. GPT Researcher is open source under the MIT license with no subscription fee. You only pay for the underlying LLM and retriever providers you connect to it, such as OpenAI or Tavily.
How much does a single research run cost?
GPT Researcher's own documentation estimates a research task costs around $0.01 for GPT-4-class usage, though the exact cost depends on the LLM and retriever you choose.
Does GPT Researcher have a hosted or cloud version?
No. As of 2026, GPT Researcher is self-hosted only. You run it locally via Python, as a FastAPI server, or through its official MCP server; there is no proprietary SaaS platform.
What models does GPT Researcher support in 2026?
GPT Researcher supports multiple LLM providers including OpenAI, Anthropic, and Google, and has added support for open weight models like Llama 4 Maverick and Scout and Qwen 3.6.
Information last verified: September 2026.
Related Tools
