ScienceSwarm
Free online forum where researchers and AI agents coordinate on open math and science problems.
ScienceSwarm curates public problems that Terence Tao's wiki and ErdosProblems.com already track for free, while DeepMind and OpenAI solve the same problems without it.
ScienceSwarm is a free website where scientists can team up to tackle unsolved math and science problems together, with AI tools helping along the way. The problem is that Google DeepMind and OpenAI are already solving these exact same problems on their own, without needing a platform or a community, they just build better AI and publish the answers. If DeepMind or OpenAI announce they have cracked a major unsolved problem in the next six months, it will show that big labs do not need ScienceSwarm's help, which would make the whole platform much harder to justify to scientists.
ScienceSwarm reports repeat researcher engagement or problem claims within 6 months of launch. Would indicate real workflow lock-in rather than just signups.
ScienceSwarm announces a funding round after being absent from YC Demo Day's standout list. Would show investors validated the company despite no revenue or traction.
- DeepMind's Aletheia solved 4 Erdos problems autonomously
- Absent from TechCrunch's 11 standout YC Demo Day startups
- Platform remains free with no reported revenue or customers
- No revenue model or pricing; platform is entirely free with no monetization path disclosed.
- Zero proprietary data, integration, regulatory, lock-in, or distribution moats identified.
- Frontier labs solve the identical public problem dataset directly, making community layer redundant.
- Not named among YC's Demo Day standout startups as of June 18, 2026, signaling weak investor pull.
- Inferred single-digit headcount competing against far better resourced incumbents.
| Rival | Direction | Evidence |
|---|---|---|
| Google DeepMind | losing to | Aletheia agent solved 4 Erdos problems directly, no platform needed, Feb 2026 |
| OpenAI | losing to | Internal model solved 80-year-old Erdos Unit Distance Problem, reported May 22, 2026 |
| Terence Tao's community wiki | losing to | Free, non-commercial tracker of AI Erdos progress, noted March 2, 2026 |
| Kaggle (Google-owned) | contesting | Could add a science-problem category using existing Google infrastructure at any time |
- Is ScienceSwarm profitable in 2026?
- No. ScienceSwarm is a free, sign-up-and-use platform with no disclosed revenue, pricing, or customer base as of its March 12, 2026 YC launch. The founders' public ask was for user introductions, not customers, and no funding amount has been confirmed publicly.
- Could AI replace ScienceSwarm's core function?
- Largely yes. Google DeepMind's Aletheia agent autonomously solved four open Erdős problems around February 2026, and OpenAI's internal model produced a new Erdős Unit Distance result by May 22, 2026, both bypassing any community platform like ScienceSwarm entirely.
- Who founded ScienceSwarm and who owns it?
- ScienceSwarm is a private, YC-backed startup founded by two named cofounders, Peter and Seiji, which launched publicly via Y Combinator on March 12, 2026. It has no confirmed external investor names or valuation disclosed in public sources as of July 27, 2026.
- Who is ScienceSwarm's biggest competitor?
- Google DeepMind and OpenAI are the most direct threats. Both deployed frontier agents straight against the same public Erdős problem dataset ScienceSwarm curates, with DeepMind's Aletheia solving four problems around February 2026 and OpenAI following in May 2026, without needing any community layer.
- Does ScienceSwarm have paying customers or revenue?
- No revenue figures, customer counts, or enterprise logos have been found anywhere. At its March 12, 2026 launch, founders described target users as researchers, PhD students, citizen scientists and agent builders, and the platform remains free to sign up and use.
- What problems does ScienceSwarm host on its platform?
- ScienceSwarm seeded its platform at launch, March 12, 2026, with thousands of open problems, explicitly including the Erdős problem collection and the Millennium Prize Problems, for AI agents and human researchers to attempt collaboratively.
- 2026-06-18TechCrunch's list of 11 standout YC Demo Day startups does not include ScienceSwarm
- 2026-06-07arXiv paper documents LLM proof search against 353 Erdos problems and 492 OEIS conjectures
- 2026-05-22OpenAI's internal model reportedly solves 80-year-old Erdos Unit Distance Problem
- 2026-03-12ScienceSwarm publicly launches via Y Combinator with founders Peter and Seiji
- 2026-03-09Report surfaces that DeepMind's Aletheia agent autonomously solved 4 Erdos problems
- 2026-03-02Terence Tao's community wiki noted as tracking AI contributions to Erdos problems for free
- 2026-02Google DeepMind's Aletheia agent autonomously solves four open Erdos problems
- Launch YC: ScienceSwarm 🔬 Doing for Science What AI Coding Did for Software (2026-03-12)
- Aletheia: Google DeepMind's AI Just Solved 4 Erdős Problems Autonomously (2026-03-09)
- AI just solved an 80-year-old 'Erdős problem,' and mathematicians are amazed (2026-05-22)
- Advancing Mathematics Research with AI-Driven Formal Proof Search (2026-06-07)
- The 11 standout startups from YC's Demo Day, according to VCs (2026-06-18)
- Erdős Problems, Combinatorics Search, and the Benchmark Frontier (2026-03-02)
- ScienceSwarm 2026 Company Profile (2026-07-27)
- Search Startups funded by Y Combinator (YC) 2026 (2026-07-25)
- Kaggle