Vending-Bench 2 (leaderboard)

Andon Labs' leaderboard for Vending-Bench 2, in which a model runs a simulated vending business for a year from a $500 start. As read on 19 October 2026, the best model, GPT-6 Astra, averaged $15,514.70 (± $1,074). Andon estimates that a good strategy could make roughly $63k in a year, and fits the frontier rising by about $822 a month. A year's run takes 3,000 to 6,000 messages. The site cites it as a simulated test of an agent running a business on its own.

Cited in

a method page on this site (Method: when could an agent earn its keep alone?)

Used for

Earning; running a business alone

Strand

The agent economy · Agents earning, working & fundraising

Cite

@misc{andonlabs2026vendingbench2,
  title = {Vending-Bench 2 (leaderboard)},
  author = {Andon Labs},
  year = {2026},
  howpublished = {Andon Labs},
  url = {https://andonlabs.com/evals/vending-bench-2},
}