AI-Researcher

tom@tomrochette.com (Tom Rochette) — Sun, 11 Oct 2026 00:00:00 +0000

AI-Researcher is HKU Data Intelligence Lab’s fully autonomous research system that takes a research idea through literature review, algorithm implementation, and publication-ready manuscript writing, and its NeurIPS 2025 paper shipped with Scientist-Bench, the team’s own benchmark for scoring such systems.

It is the idea-to-paper sub-family’s largest community and its least reusable artifact: 5,802 GitHub stars against a repository with no license file, dormant since October 2025, with the project’s momentum moved into a commercial product the repository only links.

What it is
#

A Python system (github.com/HKUDS/AI-Researcher, open-sourced May 11, 2025) by Jiabin Tang, Lianghao Xia, Zhonghang Li, and advisor Chao Huang of HKU, with the paper (arXiv 2505.18705) accepted at NeurIPS 2025. Input arrives at two levels: a detailed idea description the system develops as specified, or a brief research direction it expands itself. The paper describes orchestration of the complete pipeline from literature review and hypothesis generation to algorithm implementation and publication-ready manuscript preparation, evaluated on Scientist-Bench, a benchmark of state-of-the-art papers across AI domains with guided-innovation and open-ended tracks. The GitHub description points to novix.science/chat as the production-ready version.

Status
#

Dormant as an open repository, carrying the sub-family’s biggest audience: 5,802 stars and 725 forks, created 2025-03-11, last pushed 2025-10-16, as of 2026-10-11. No relevant Hacker News threads surfaced in this run’s search; the footprint is GitHub-native, with a Trendshift badge and Slack and Discord communities.

Strengths
#

Scale of validation interest: the sub-family’s largest star and fork counts, a NeurIPS 2025 acceptance, and a Trendshift badge.
Scientist-Bench gives the sub-family something it lacked: a named benchmark with guided and open-ended innovation tracks, published as a leaderboard.
The two-level input design matches how research actually starts, from a formed idea or a vague direction.
A clear commercial continuation (Novix) rather than plain abandonment.

Cautions
#

There is no license file in the repository: the code can be read but not legally reused, which matters for the sub-family’s most-starred repository.
Dormant since 2025-10-16, so its integrations are nearly a year stale.
Scientist-Bench is the team’s own benchmark, self-run and self-scored; no independent evaluation surfaced in this run’s searches.
The commercial Novix product is where development went, so the open repository functions as the paper’s artifact rather than a maintained tool.

Pricing
#

Free to run as code, with no license permitting reuse; there is no published paid tier, so pricing does not apply beyond your own API keys. The Novix product’s prices are not published in the sources fetched this run.

Compared to
#

Sakana AI Scientist: the sub-family ancestor with the peer-review milestone; AI-Researcher has the larger community, the license gap, and the benchmark contribution.
AutoResearchClaw: the actively maintained pipeline with citation verification and human-oversight modes; choose it over AI-Researcher unless you need Scientist-Bench.
Agent Laboratory: the human-assist ancestor with the candid quality measurements; AI-Researcher’s evaluations are its own.

Bottom line
#

Recommended reading for Scientist-Bench and as a case study in what a big star count without a license is worth: admiration without reuse. Not for anyone needing a runnable, maintained system: the repository is a year dormant and legally closed.

Changes
#

2026-10-11 - Created.

References
#

https://github.com/HKUDS/AI-Researcher - repository, 5,802 stars, NeurIPS 2025 description, last push 2025-10-16, no license detected (as of 2026-10-11)
https://raw.githubusercontent.com/HKUDS/AI-Researcher/main/README.md - the two-level input design, communities, and production-version pointer (fetched 200, 2026-10-11)
https://arxiv.org/abs/2505.18705 - the paper: authors, pipeline scope, and Scientist-Bench (fetched 200 via the arXiv API, 2026-10-11)
https://autoresearcher.github.io - the HKU team, the May 11 2025 open-source release, and the Scientist-Bench leaderboard (fetched 200, 2026-10-11)
https://api.github.com/repos/HKUDS/AI-Researcher/contents/ - the root file listing confirming no LICENSE file (fetched 200, 2026-10-11)
https://hn.algolia.com/api/v1/search?query=%22AI-Researcher%22 - the search returning no relevant threads (fetched 200, 2026-10-11)

hkuds on tomrochette.com

AI-Researcher

What it is
#

Status
#

Strengths
#

Cautions
#

Pricing
#

Compared to
#

Bottom line
#

Changes
#

See also
#

References
#

hkuds on tomrochette.com

AI-Researcher

What it is #

Status #

Strengths #

Cautions #

Pricing #

Compared to #

Bottom line #

Changes #

See also #

References #

What it is
#

Status
#

Strengths
#

Cautions
#

Pricing
#

Compared to
#

Bottom line
#

Changes
#

See also
#

References
#