benchmarks.site

Public leaderboard · Agents, tool use & computer use

Vending-Bench 2

An AI runs a simulated vending-machine business for a long time; tests staying sensible over many days.

Leaderboard

Top 3 right now

  1. 1stGemini 4 Carbon$16,484.21 ± $1,096
  2. 2ndGPT-6 Astra$15,514.70 ± $1,074
  3. 3rdGemini 4 Hydrogen$14,549.49 ± $495

What it measures

An agent runs a simulated vending-machine business for a simulated year (suppliers, prices, refunds, stock), scored by its money balance at the end.

How to read the numbers

Average final money balance ($) — higher is better.

Results

10 rows · higher is better (Final money balance)CSV
1Gemini 4 Carbon$16,484.21 ± $1,096
2GPT-6 Astra$15,514.70 ± $1,074
3Gemini 4 Hydrogen$14,549.49 ± $495
4GPT-6 Sol$14,427.85 ± $1,051
5Gemini 4 Argon$13,718.16 ± $3,100
6GPT-6.1 Sol$12,860.61 ± $1,043
7Muse Wonder Full$12,291.56 ± $885
8Claude Opus 5$11,181.87 ± $2,094
9Claude Sonnet 5.5$11,165.28 ± $1,585
10Claude Opus 4.7$10,936.76 ± $1,181

Note: Top 10 of 76 shown on page; no update date shown, but includes very recent models.

Source: Andon Labs · andonlabs.com · retrieved 2026-11-17

Markdown · More Agents, tool use & computer use