Public Goods Game Average Contribution

Name: Public Goods Game Average Contribution
Published: 2025-02-22

Games · 2025-02-22

Lech Mazur's Public Goods Game benchmark measures how much models contribute in a repeated public-goods scenario. The public README reports average percentage contribution with standard errors; because contribution is a behavioral preference rather than a monotonic capability outcome, Atlas retains it as a diagnostic instead of fitting it as capability evidence.

Top models (higher is better)

Model	Score
Gemini 2.0 Flash Experimental	45.2
Haiku 3.5	41.0
Gemma 2 27B IT	31.2
Gemini 1.5 Flash (source unspecified)	30.2
GPT-4o Mini	24.8
Gemini 1.5 Pro 002	24.6
Qwen2.5 72B Instruct	23.6
Claude 3.5 Sonnet (Oct 2024)	20.9
Gemini 2.0 Flash Thinking Experimental 01-21	20.0
Grok 2	15.8
DeepSeek-V3	15.5
GPT-4o (source unspecified)	14.7
Llama 3.1 405B Instruct	3.2
o1-mini	2.6
o1 (source unspecified)	2.5
Mistral Large 2 (Instruct 2407)	1.6
Llama 3.3 70B Instruct	1.2
Qwen2.5-Max	0.5
DeepSeek-R1	0.3
o3-mini	0.0