Was OpenAI’s Millennium Prize problem solution primarily achieved using a large multi‑agent swarm of collaborating models?
The dispute is whether OpenAI’s Millennium Prize problem solution came mainly from a huge swarm of agents or from one powerful model. Soares (The Diary Of A CEO), Rodriguez (80,000 Hours), and Ladish (The Diary Of A CEO) say a swarm of 10,000 agents solved a Millennium problem in about a week. Brown (Dwarkesh Podcast) and Shapira (Doom Debates) say the key advance over the past two months was a single very strong model, with multi-agent systems secondary.
The solution mainly came from a large swarm of collaborating OpenAI agents.
Soares (The Diary Of A CEO) says OpenAI ran 10,000 agents for 11 days to solve a Millennium problem, and this capability existed at least six months earlier. Rodriguez (80,000 Hours) says a swarm of OpenAI models worked together over a week to produce a solution to a Millennium Prize Problem. Rodriguez adds that this swarm solved an open math problem that world-class mathematicians had worked on for over 90 years. Ladish (The Diary Of A CEO) says 10,000 OpenAI agents collaboratively solved a decades-old Millennium math problem after OpenAI achieved success training agents to work together earlier that year.
4 claims, in their words
-
“it was 10,000 agents running for 11 days. And they definitely could have done that six months ago.”
-
“When GPT-4 was released in 2023, it could hardly multiply two numbers together. But within the last few weeks, a swarm of OpenAI models worked together to produce a solution to the Millennium Prize Problem”
-
“a swarm of OpenAI models worked together to produce a solution to the Millennium Prize Problem, an open math problem that world-class mathematicians have been working on for over 90 years. The AIs did it in a week.”
-
“10,000 agents from OpenAI worked together to solve a millennium problem. This is one of the hardest problems in mathematics. It's been open for decades. ... OpenAI said they didn't have success at training agents to work together until this year.”
The solution was mainly produced by a single powerful model, not multi‑agent swarms.
Brown (Dwarkesh Podcast) says the Millennium Prize solution was not due to multi-agent methods and assigns under 10% of the credit to them. Brown says the core reason for the solution is that OpenAI now has a very powerful model. Shapira (Doom Debates) says OpenAI let a model built for software engineering spend a few hours solving a Millennium Prize problem as a side project. Shapira says OpenAI described this as not their focus and as using a model built for software engineering customers, not a bespoke multi-agent research system.
2 claims, in their words
-
“This was not due to multi-agent. I wouldn't even like attribute like 10% of the credits to multi-agent. ... But the core reason is like, this is just a very powerful model.”
-
“OpenAI was like, yeah, you know, this isn't really our focus. We just decided, hey, why don't we just take this model that we're making to sell to software engineering companies? Why don't we just give it a few hours to go solve a Millennium prize problem? Oh, look, it did it.”
Listeners of “The solution mainly came from a large swarm of collaborating OpenAI agents.” have not heard
Listeners of these shows have not heard Brown’s point that the Millennium Prize solution was mainly due to a single powerful model, with multi-agent systems deserving under 10% credit.
1 unanswered argument, in their words
-
“This was not due to multi-agent. I wouldn't even like attribute like 10% of the credits to multi-agent. ... But the core reason is like, this is just a very powerful model.”
Listeners of “The solution was mainly produced by a single powerful model, not multi‑agent swarms.” have not heard
Listeners of these shows have not heard Soares’s description of OpenAI running 10,000 agents for 11 days to solve a Millennium problem, a setup possible at least six months earlier.
1 unanswered argument, in their words
-
“it was 10,000 agents running for 11 days. And they definitely could have done that six months ago.”
The swarm side leans on reports of 10,000 agents working a week on a decades-old Millennium problem, while the single-model side emphasizes a powerful model. One narrative treats multi-agent coordination as central to the achievement, while the other treats it as incidental and assigns it under 10% of the credit.
- Nov 14, 2026 Liron Shapira (Doom Debates, 2026-10-02) addressed an argument by Nate Soares.
- Nov 14, 2026 3 new claims (Dwarkesh Podcast).
- Nov 13, 2026 Published: cleared the gate.
- Nov 13, 2026 7 new claims (Doom Debates, Dwarkesh Podcast, The Diary Of A CEO, 80,000 Hours Podcast).
Published because: 4 shows, 2 positions, 21 days of debate. Questions publish only when claims come from at least three shows, at least two positions are held by different people, and the debate has lasted two weeks.
Pairs are matched by content across different episodes and shows; no exchange between these speakers took place unless both quotes come from the same episode.
The prose on this page is written by a model from the quotes shown, and every sentence of it rests on quotes you can open; sentences that could not be traced to a quote were removed before publishing. Positions are placed from each speaker's own words. Evidence labels say what the speaker cited, never whether it is true.
7 episodes behind this page
- 2026-11-13Doom Debates — Jensen Huang Resigns, RSI Coming in 2027 — with Robert Wright
- 2026-11-13Dwarkesh Podcast — The avalanche that hit math is coming for AI development
- 2026-10-08The Diary Of A CEO — AI Safety Whistleblower: 700 AI Agents Attacked A Company To Cover Their Tracks! | Jeffrey Ladish
- 2026-10-02Doom Debates — Trump and the AI Safety Vibe Shift, With Robert Wright | Nonzero × Doom Debates
- 2026-09-2480,000 Hours Podcast — How we get from AI cyberattacks to human extinction
- 2026-09-17The Diary Of A CEO — The Great AI Debate: Is Artificial Intelligence an Extinction Threat? Debating the True Risks of Advanced Models
- 2026-09-17Dwarkesh Podcast — Noam Brown – Agent swarms, alignment, & recursive self-improvement
The week's bottom lines, in your inbox
One email a week: the stories that mattered and what they actually mean. Free.