Benchmarks / Agentic

Vending-Bench Arena

Multi-agent Vending-Bench 2: models run competing vending machines at one location, may trade or collude, and are scored individually on final balance.

12models
–Open Index weight
–last updated
Andon Labspublisher

Vending-Bench Arena

Higher is better · best effort per model

Chart settings
12 of 12 selected

full table below

#ModelEffortScoreCost / taskHarness
1GPT-6 Astramax12400.00 USD––
2Opus 5.5max8100.00 USD––
3GPT-5.5xhigh7980.00 USD––
4GLM 5.3max7800.00 USD––
5GPT-5.6 Solmax7400.00 USD––
6Opus 5max7000.00 USD––
7Opus 4.7max5838.00 USD––
8Sonnet 4.6max5639.00 USD––
9Opus 4.8max5188.00 USD––
10Fable 5max4200.00 USD––
11Opus 4.6max4053.00 USD––
12Kimi K3max3200.00 USD––