Atlas

Benchmarks

← All benchmarks

Vending-Bench

Games · 2025-02-20

Vending-Bench is a long-horizon business simulation where agents operate a vending-machine business over many turns; reported metrics include final valuation and units sold.

Top models (higher is better)

ModelScore
Grok 44694
Gemini 3 Pro Preview4388
GPT-53579
Sonnet 4.52465
GPT-5.12380
Claude 3.5 Sonnet (Oct 2024)2218
Opus 42077
o31843
Claude 3.7 Sonnet1568
Kimi K2 Thinking1296
Gemini Robotics-ER 1.5976
Sonnet 4968
o3-mini907
Human (Vending-Bench)844
Gemini 2.5 Pro Preview 03-25789
Qwen3-235B-A22B616
gpt-oss-120b608
Gemini 1.5 Pro594
GPT-4o Mini582
GPT-5 Mini581
Gemini 1.5 Flash572
Haiku 3.5373
Gemini 2.0 Flash338
GPT-4o335
Gemini 2.0 Pro Experimental274
Llama 4 Maverick157
Loading Atlas data…