- Cleus
- AI benchmarks
- Vending-Bench 2
Vending-Bench 2explained
Can it run a small business for a year on its own?. The model runs a simulated vending-machine business for a simulated year. It starts with $500 and has to find suppliers online, order and haggle by email, stock the machine, set prices and pay a $2 daily fee. Some suppliers try to cheat it. The score is how much money it ends up with.
- Highest published score
- Gemini 4 Carbon, $16,484
- Leader among popular models
- GPT-6 Astra
- Models with a result
- 65
- Skill it tests
- Agents
What Vending-Bench 2 really tests
- What it is
- The model runs a simulated vending-machine business for a simulated year. It starts with $500 and has to find suppliers online, order and haggle by email, stock the machine, set prices and pay a $2 daily fee. Some suppliers try to cheat it. The score is how much money it ends up with.
- A question like
- A supplier quotes too high, a product stops selling in winter and a delivery arrives short. Does the model notice, push back and adjust, or drift into bankruptcy? Our own example in the style of the test, not a real item.
- Why it matters
- It tests staying sensible over thousands of steps, which is what you need from an AI assistant that acts for you over days or weeks.
- Reading the score
- Money in the bank at the end of the simulated year. The model starts with $500.
- The catch
- It is one simulated business, and results can swing between runs. A big balance means a model was consistently sensible, not that it would run your company.
- Made by
- Andon Labs, 2025. Official page
Leaderboard
Each model’s best published result across its reasoning settings.
Popular models
The models we track across this site.
Top 20 of all 65 models tested
Includes older, preview and specialist models that aren’t on our main leaderboard.
Questions
What is Vending-Bench 2?
The model runs a simulated vending-machine business for a simulated year. It starts with $500 and has to find suppliers online, order and haggle by email, stock the machine, set prices and pay a $2 daily fee. Some suppliers try to cheat it. The score is how much money it ends up with.
What does a Vending-Bench 2 score mean?
Money in the bank at the end of the simulated year. The model starts with $500.
Which AI model is best on Vending-Bench 2?
Gemini 4 Carbon from Google DeepMind has the highest published Vending-Bench 2 result we have ($16,484). Among the popular models on this site, GPT-6 Astra leads with $15,515.
Is Vending-Bench 2 a good benchmark?
It tests staying sensible over thousands of steps, which is what you need from an AI assistant that acts for you over days or weeks. The catch: It is one simulated business, and results can swing between runs. A big balance means a model was consistently sensible, not that it would run your company.
Who made Vending-Bench 2?
Andon Labs, 2025.
Benchmark scores: Epoch AI, Capabilities & benchmarking, used under CC BY 4.0. For each model we show its best published result across reasoning settings. Prices and context sizes: OpenRouter. Data refreshed November 9, 2026. Hands-on tests (answers, websites, code, refusals, speed) are our own single runs. Explanations written by Cleus.