GPT Researcher

Overview

GPT Researcher is the original open-source deep research agent — an autonomous system that produces detailed, factual, and unbiased research reports with citations. Created in May 2023 by Assaf Elovic, it predates the deep research wave by nearly two years and remains the most-starred open-source deep research tool on GitHub at 27.6k stars.

The core architecture uses a planner/execution pattern: the planner generates research questions from a user query, crawler agents gather information from 20+ web sources in parallel, and a publisher aggregates all findings into a comprehensive report with source tracking.

Key stats (as of June 11, 2026): 27,643 stars, 3,721 forks, Apache-2.0 license, Python, active development since May 2023 — last push May 28, 2026.

---

Status & Releases

Development remains active. Since this profile's original date (March 2026), the project has shipped v3.4.4 (April 16, 2026) and v3.5.0 (May 28, 2026), continuing a steady monthly-ish release cadence that goes back to 2023. The repo is not archived and carries 218 open issues — normal load for a project this size.

Beyond standard web research, the docs cover multi-agent research teams (planning through publication), hybrid web + local document research, and a recursive Deep Research mode.

---

Architecture

The agent follows a structured pipeline:

1. Create a task-specific agent based on the research query
2. Generate questions that collectively form an objective opinion on the task
3. Use crawler agents to gather information for each question in parallel
4. Summarize and source-track each resource
5. Filter and aggregate summaries into a final research report

The parallelized approach addresses key problems with naive LLM research: it reduces hallucination by cross-referencing multiple sources, increases speed through concurrent crawling, and mitigates bias by drawing from diverse sources.

---

Pricing & Licensing

There is no commercial hosted tier as of June 2026 — GPT Researcher is a self-hosted tool, not a SaaS.

---

Competitive Position

Strengths: Longest-running deep research agent. Battle-tested, well-documented, large community. Apache-2.0 license. Works with any LLM provider. Claude Code skill integration.

Weaknesses: Prompt-based approach (no fine-tuned model). Newer tools like Tongyi DeepResearch are beating it on benchmarks with RL-trained specialized models.

---

Cautions

• Report quality is bounded by the underlying model and search provider you configure — it's an orchestration layer, not a trained research model
• Per-run API costs (~$0.10 average) scale with depth settings; recursive Deep Research runs cost more
• The prompt-based architecture is losing benchmark ground to RL-trained agents, so expect the gap to widen on hard research tasks

---

What Developers Say

Community sentiment on Hacker News is modest in volume but consistently positive, centering on control and configurability. Comparing it to proprietary deep research products, sublimefire wrote: "I do prefer tools like GPT researcher where you are in control over sources and search engines." On cost flexibility, jimmySixDOF noted "there are so many ways to configure GPT Researcher for all kinds of budgets." On maintenance, wluk observed: "The creator of the project, Assaf, has been nothing but friendly." Notably, the project has never had a large standalone HN thread — its 27k-star adoption grew through GitHub and word of mouth rather than a viral launch.

---

Bottom Line

Recommended if you want a self-hosted, provider-agnostic deep research agent with full control over sources, search engines, and cost — it's the most mature open option, still actively shipping (v3.5.0, May 2026).

Not recommended if you need state-of-the-art benchmark performance on hard research tasks — RL-trained agents like Tongyi DeepResearch now lead there — or if you want a managed SaaS with zero setup.

Outlook: Stable and durable. Three years of continuous releases and 27.6k stars make it the default open-source baseline for the category, even as fine-tuned challengers pass it on benchmarks.

---

Research by Ry Walker Research • methodology