Atlas

Benchmarks

← All benchmarks

Vending-Bench (sales-duration percentage)

Games · 2025-02-20

Percentage of the Vending-Bench simulation completed before sales stopped, as published by Andon Labs.

Top models (higher is better)

ModelScore
GPT-5100.0
Human (Vending-Bench)100.0
GPT-5.1100.0
Opus 499.5
Grok 499.5
Gemini 3 Pro Preview94.6
GPT-5 Mini92.8
gpt-oss-120b92.7
Gemini Robotics-ER 1.591.1
Gemini 2.5 Pro Preview 03-2589.0
o386.9
Claude 3.5 Sonnet (Oct 2024)82.2
Sonnet 480.6
Claude 3.7 Sonnet80.3
o3-mini80.3
Sonnet 4.577.8
Kimi K2 Thinking75.4
Qwen3-235B-A22B73.2
GPT-4o Mini73.2
Gemini 2.0 Flash55.7
GPT-4o50.3
Gemini 1.5 Pro43.8
Gemini 1.5 Flash42.4
Gemini 2.0 Pro Experimental15.8
Haiku 3.512.9
Llama 4 Maverick1.2
Loading Atlas data…