Not evaluated
EinsteinArena: AI agents on open problems with public verifiers
EinsteinArena: a platform where AI agents tackle open problems with public verifiers
The gist
EinsteinArena is a platform where AI agents work on open mathematics problems with public verifiers; Together AI and Stanford announce new records (blog post of April 13, 2026); no DiscoverIA status yet.
DiscoverIA status on October 12, 2026: not evaluated, meaning no third-party evaluation found.
AI draft, not reviewedWhat has not been reviewed?
Text written by an AI (DiscoverIA pipeline), on . The status and its sources come from the registry; only the explanatory text has not been reviewed. A reported error is corrected, then the text is reviewed.
Cite this entry
A single status snapshot since Oct 12, 2026: a citation always points to a dated snapshot.
Cite snapshot 1, Oct 12, 2026
APA citation with the dated link
BibTeX @misc, key discoveria-einsteinarena-together-ai-r1
@misc{discoveria-einsteinarena-together-ai-r1,
author = {{DiscoverIA}},
title = {EinsteinArena: a platform where AI agents tackle open problems with public verifiers},
howpublished = {DiscoverIA registry, entry einsteinarena-together-ai},
note = {Status “Not evaluated”, snapshot 1, recorded on October 12, 2026. New result},
year = {2026},
url = {https://discoveria.ai/entry/einsteinarena-together-ai#releve-1},
urldate = {2026-10-12}
}Dated link opens this snapshot in the history
To understand ≈ 30 s reading time, estimated
What it changes
If the records hold, this kind of platform would let anyone follow in public, problem by problem, how far AI agents push known values, each with its verifier and a downloadable solution. The batch gathers 16 entries, all not yet evaluated. For the kissing number in dimension 11, the paper moves the bound from 593 to 604.
What remains unclear
The verifiers are programs run by the platform, not third parties: no outside reviewer is named. The count of records differs by source: 12 on the arXiv abstract page, 11 in the HTML abstract, in table 1 and in the blog, which makes 13 problems when the paper and the blog are combined. The paper states that the result at 604 came from a series of submissions, public discussions and verifier revisions.
In the glossary : Agent
Read the paper
Harnessing the Collective Intelligence of AI Agents in the Wild for New Discoveries
What it establishes : The authors present EinsteinArena and report that by May 2026 its agents had set 11 new records across 15 open problems, including the kissing number in dimension 11, raised from 593 to 604.
Where to start : The abstract, then Table 1 (Section 2.2): the 15 problems, with the previous and new records. Section 3 and Appendix B cover dimension 11 in detail, Appendix A each problem. The arXiv abstract page announces 12 records, the paper itself 11.
Written by an AI · reviewed by a human
For experts
Sources, method, key facts, status history, report an error
Status
Not evaluated
Recorded on · together.ai
Step 1 of 4 on the scale: not evaluated. Next step: under review.
- Announced
- Recordednot evaluated
- nextTo follow
What this status means
Source reading : Source read word for word (level 3 of 3)
The result is declared neither right nor wrong.
To move to “Under review”: a public review under way. To “Peer reviewed”: a journal publication or a review by named experts.
Sources
- Main sourcearxiv.org · 2606.10402v2page read word for word · consulted on Oct 12, 2026 ·
improved the best known lower bound from 593 to 604
- Main sourcearxiv.org · 2606.10402page read word for word · consulted on Oct 12, 2026 ·
12 new state-of-the-art results
- Papertogether.aipage read word for word · consulted on Oct 12, 2026 ·
have already discovered the new best solutions to 11 open math problems
- Repositoryeinsteinarena.comlink only, reading not recorded
- Repositoryeinsteinarena.com · skill.mdlink only, reading not recorded
- Repositorygithub.com · EinsteinArena-new-SOTAlink only, reading not recorded
- Repositorygithub.com · einstein-arenalink only, reading not recorded
Method
DiscoverIA status: not evaluated, meaning that no third-party evaluation has been found in the sources read. Set on October 12, 2026 from together.ai (page read, quotation checked word for word: “have already discovered the new best solutions to 11 open math problems”).
Limits
No named third party: the recorded checks come from the claimant or from a platform tool. DiscoverIA has not checked this result itself.
Key facts
- Claimed by
- Together AI / Stanford (EinsteinArena)
announced on Apr 13, 2026 - Nature
- new result
- Field
- Mathematics (optimisation, discrete geometry, analysis, combinatorics (open problems with a public checker))
- Checked by
- the platform's public checkers (a tool); no third party named
- Sources
- 7 : 3 read word for word, 4 links only
- This entry
- AI draft, not reviewed
Status history
Only one observation so far.
- 01
First observation:Not evaluatedCurrent status
not evaluated: only the platform's public checkers, no third party named
Source: together.airead word for word
An observation is added; none is ever erased.
Parts of this announcement (16)
Report an error or suggest a source
Thank you, your message was sent.
Sending failed. Try again, or write to contact@discoveria.ai.
Is your result listed?What each status requires, correcting this entry, replying to it.