GPT Researcher changelog: what's new each month
Every stable GPT Researcher release summarised by month: the highlights, new features, improvements, fixes and anything you need to act on. 3 months covered; the current month updates daily.
September 2026
1 release: 3.7.0
September brought significant infrastructure improvements and a shift toward modern Python versions. The month focused on enhancing research quality, fixing multi-agent behavior, and reducing external dependencies.
- Python 3.12+ is now required; earlier versions are no longer supported.
Highlights
- Python 3.12+ is now required.
- Default context filter now uses keyword ranking instead of TypeSafe when no key is provided.
- Multi-agent sections no longer repeat each other's content.
- Frontend now uses self-hosted assets instead of third-party CDNs.
- Retriever plugins can now be distributed as separate packages.
New
- Support for distributing custom retriever plugins via the gptresearcher.retrievers entry point.
- Self-hosted frontend assets, eliminating reliance on third-party CDNs.
Improved
- Default context filtering behavior now uses keyword ranking without requiring a TypeSafe key.
- Multi-agent research no longer produces duplicate content across sections.
- 25+ community fixes across retrievers, scrapers and deep research.
Fixed
- Frontend WebSocket opening a second connection on new research runs.
August 2026
1 release: 3.6.1
August brought stability improvements and observability enhancements to GPT Researcher. The release restored critical functionality and integrated new monitoring capabilities.
Highlights
- Added agent observability with Monocle integration for better monitoring and debugging
- Restored main branch functionality after resolving 97 queued changes
- Fixed web scraper to preserve extracted content even when enrichment fails
- Documented scholarly retriever configuration for easier setup
New
- Agent observability with Monocle
Improved
- Web scraper now keeps extracted content when webbaseloader enrichment fetch fails
- Empty source URLs notification made reachable
- Scholarly retriever configuration now documented
- Python dependency floor testing added to CI
Fixed
- Restored broken star history chart in documentation
- Fixed main branch after queued pull requests
- Corrected empty-source-urls notification accessibility
- Updated python-pptx requirement to >=1.0.2
July 2026
1 release: 3.6.0
July brought significant stability and security improvements across GPT Researcher, with hardening of retrievers, scrapers, and multi-agent systems alongside new integration examples and fixed token tracking.
Highlights
- Security enhancements including content sanitization, new security policy, and dependency pinning.
- Fixed silent token limit drops and implemented accurate usage-based cost tracking.
- New Deep Agents example showing GPT Researcher integrated as a research engine in LangChain workflows.
- Comprehensive retriever and scraper robustness improvements preventing malformed data handling.
- Multi-agent revision loop bounds and exact sentinels to prevent runaway behaviors.
New
- Deep Agents example demonstrating GPT Researcher as a research engine within LangChain deep agents.
Improved
- Token limit handling now properly tracked instead of silently dropped.
- Cost tracking switched to accurate usage-based measurement.
- Research pipeline robustness across multiple components.
- Retriever guard fixes against 20 types of malformed results.
- Scraper handling of titles, PDF detection, temporary files, and image dimensions.
- Multi-agent revision loops now bounded with exact sentinels.
- Core systems hardening for costs, queries, LLM calls, agent behavior, MCP, and configuration.
Fixed
- Post-merge conflicts in retriever and context handling.
- Silent dropping of token limits in research operations.
- Scraper issues with title extraction, PDF detection, temp file cleanup, and image dimensions.
- Malformed retriever results causing pipeline failures.
- Multi-agent revision loops entering infinite states.
- Cost calculation and LLM usage tracking errors.
Summaries are written automatically from the official release notes (full changelog ↗); check the original notes before relying on a detail. GPT Researcher: pricing, features and alternatives · All changelogs