www.lesswrong.com/posts/6cb7qd3RSkgnviCpf/swarm-scaling
2 corrections found
in which time they sent 5 million messages to each other and used 300 billion tokens.
Those totals are for OpenAI’s entire multi-problem effort, not for the Navier–Stokes run itself. OpenAI says the Navier–Stokes run used about 2.7 million messages and 130 billion output tokens.
Full reasoning
OpenAI’s own writeup distinguishes between all attempted problems and the specific Navier–Stokes run.
- It says the agents reached the Navier–Stokes result about 88 hours after launch.
- But it then says "Across all attempted problems" the agents sent 4.9 million messages and used about 300 billion output tokens.
- For Navier–Stokes specifically, OpenAI says the agents sent 2.7 million messages and used approximately 130 billion output tokens.
So the post mixes the correct 88-hour figure for the Navier–Stokes run with message/token totals that OpenAI attributes to the broader overall effort, not to that run alone.
1 source
- On the Navier–Stokes Millennium Prize Problem | OpenAI
The agents arrived at their resolution on Saturday, September 5, about 88 hours after the first agents were launched... Across all attempted problems, the agents sent 4.9 million messages and used about 300 billion output tokens. In the process of resolving the Navier–Stokes problem, the agents sent 2.7 million messages and used approximately 130 billion output tokens.
the swarm of 10,000 agents cost about $20 million at API prices.
That multi-million-dollar estimate refers to OpenAI’s whole week-long 300B-token effort, not specifically to the 10,000-agent Navier–Stokes swarm. OpenAI says the Navier–Stokes run itself used about 130B output tokens, less than half of the total.
Full reasoning
The cited cost estimate is being attached to the wrong scope.
TechCrunch’s estimate says "All told, the week-long effort consumed 300 billion output tokens" and prices that overall effort at $22.5 million using Astra rates. OpenAI’s own post, however, says that the Navier–Stokes run itself used approximately 130 billion output tokens, while the 300 billion total was for all attempted problems.
So the post’s wording is misleading: the ~$20M figure is an estimate for the broader multi-problem campaign, not specifically for the 10,000-agent swarm that produced the Navier–Stokes result.
2 sources
- OpenAI fought dirty on career-making math problem, says NYU mathematician | TechCrunch
All told, the week-long effort consumed 300 billion output tokens — $22.5 million worth of compute, if charged at current Astra rates.
- On the Navier–Stokes Millennium Prize Problem | OpenAI
Across all attempted problems, the agents sent 4.9 million messages and used about 300 billion output tokens. In the process of resolving the Navier–Stokes problem, the agents sent 2.7 million messages and used approximately 130 billion output tokens.