Follow every step that moves human knowledgeforward.
Track how AI and people advance questions in mathematics, physics, chemistry, life science, and beyond. Read the evidence and its limits before judging possible practical value.
Official manuscript catalogue · continuously revised
This does not mean “722 hard problems were all solved.”
The initial release contained 722 manuscripts; the current repository contains 719 manuscripts in 372 result families. This site separately marks manuscripts listed in the official Lean source catalogue and links each record.
Organized around questions, not trending topics. Choose a discipline to enter its questions directly.
AI AND THE RESEARCH PROCESS
Which specific question did AI actually advance?
Company development, academic teams, and individual or joint research are identified separately—then so are conjectures, candidates, proofs, and experimental validation.
The free public edition creates no visitor identity and keeps no cumulative visitor database.
THE QUESTIONS THAT MATTER
See the frontier through its questions.
— question guides. This is a research map; open the related cards for detailed advances, conditions, and sources.
AI × ACADEMIC PROGRESS
What did AI advance, and how far does the evidence reach?
This is not a model leaderboard. Every record corresponds to a specific research question and separates the role of AI, the role of people, verification status, and primary sources.
Inclusion thresholdA specific question must be identifiableA paper or official primary source is requiredThe unresolved remainder must be stated
BY YEAR · BY EVIDENCE
See real progress year by year.
“A company built the AI” and “researchers used AI in their work” are identified separately.
OPENAI MATHEMATICS · CHECKABLE CATALOGUE
Which of the “722 manuscripts” appear in the public Lean catalogue?
The initial release contained 722 manuscripts; the current official repository contains 719 manuscripts and 372 result families. Its README reports “300 / 719 top-line results formalized,” but that figure, the number of result families, and the number of YAML sources use different units. Frontier Atlas links each item to its exact current path and lists — manuscripts from — result families; each entry has a plain-language explanation and a link to the original manuscript.
Listed in the Lean source catalogueNot the same as final confirmation
“Finally confirmed”: no reliable public per-manuscript list currently supports such a column.The official overall status is Partial progress; review: unchecked. The current YAML lists 173 source PDFs: 172 can be matched exactly to the present result catalogue, while one was not merged automatically because its path changed after revision. Frontier Atlas has not independently compiled or audited the Lean project. Inclusion in the catalogue does not by itself confirm novelty, citation completeness, every claim in the manuscript, or completion of peer review. If official per-manuscript review results are published later, confirmed items can then be listed separately.View the official revision history ↗
MODEL & PRODUCT UPDATES
Latest AI updates.
Model and product releases are tracked separately; they do not count as “solving an academic problem.”
YOUR PERSONAL FRONTIER
Saved.
Keep the questions you want to revisit here. Saved items remain in this browser and no email is sent.
FRONTIER ATLAS · RESEARCH BRIEF · VOL. 01
October 2026 First reading brief.
Selected research advances ordered by first public release. Checked through 2026.10.13; this is an editorial reading list, not a comprehensive monthly review.
Edition assembled 2026.10.13 · Static content does not update automatically
ABOUT THE AUTHOR · RESEARCHER · ENGINEER
About the Author.
Xinjun Wang is an engineer and independent researcher whose interests span theoretical physics, quantum information, mathematics, formal reasoning, and computational structures. He initiated, designed, and continues to maintain Frontier Atlas to place new results back in the context of their questions and clearly distinguish evidence, limits, and what remains unresolved.
Frontier knowledge is scattered across papers, preprints, and institutional pages. This site helps a general reader understand in one sitting what was blocking progress, how far a result moved the question, and how strong the evidence is.
02
Authorship and editorial responsibility
Xinjun Wang is responsible for the site’s selection, structure, writing, and continuing maintenance. External paper links support source checking; inclusion does not endorse every claim in a paper.
PUBLICATIONS & DOI RECORDS
A verifiable research record.
59 DOI or journal-identifier records · 1 additional record without a DOI Current and selected historical records · Checked 2026.10.13
HOW WE READ THE FRONTIER
Make sources clear. Keep judgments bounded.
Before deciding what is “latest,” understand what kind of result it is.
01
Prefer primary sources
Papers, preprints, collaborations, standards bodies, and official repositories. Use a DOI when available and preserve the revision trail.
02
Separate levels of verification
Numerical agreement, formal proof, experimental validation, peer review, and independent replication are different and are labelled separately.
03
Separate AI and human contributions
Model developers, tool users, question setters, proof authors, and experimental validators are identified separately. System capabilities are not described as an autonomous scientist.
04
State the boundary every time
Improving a bound is not the same as resolving a complete conjecture; finding a candidate does not mean a drug or material is ready to use.
Current coverage
The site covers nine areas: mathematics, physics, computer science, chemistry, life sciences, medicine and health, Earth and space, materials and engineering, and the humanities and social sciences. The question map and AI cases continue to expand and do not claim to exhaust the research frontier.
The OpenAI mathematics catalogue is parsed from the current version of the official repository. A “result family” may include a main result, supporting argument, corollary, or alternative proof, and cannot be equated directly with the number of independently solved problems.
READ FURTHER
Continue reading and cross-checking.
External sites are provided as research entry points; their inclusion does not endorse all of their content.