244 questions answered · 244 with human benchmark data · Released 2025-08-05
gpt-oss-120b is an open-weight, 117B-parameter Mixture-of-Experts (MoE) language model from OpenAI designed for high-reasoning, agentic, and general-purpose production use cases. It activates 5.1B parameters per forward pass and is optimized to run on a single H100 GPU with native MXFP4 quantization. The model supports configurable reasoning depth, full chain-of-thought access, and native tool use, including function calling, browsing, and structured output generation.
Scores by Category
Notable Questions
Human Alignment
Would you be uncomfortable having people who have AIDS as neighbors?
No
No
Would you be uncomfortable having homosexuals as neighbors?
No
No
Wittgenstein (which do you prefer?): early or late?
Late
Late
Which statement comes closest to expressing what you believe about God?
don't believe
no doubts
God: theism or atheism?
Other
Atheism
Principle of sufficient reason: true or false?
Other
False
AI Consensus
Is it especially important that children are encouraged to learn a feeling of responsibility at home?
Yes
Yes
Is it especially important that children are encouraged to learn tolerance and respect for other people at home?
Yes
Yes
Is it especially important that children are encouraged to learn independence at home?
Yes
Yes
Personal identity: biological view, psychological view, or further-fact view?
Other
Psychological view
Arguments for theism (which is strongest?): cosmological, design, ontological, pragmatic, or moral?
Design
Cosmological
Would you be uncomfortable having drug addicts as neighbors?
Yes
No
Response Confidence
External world: idealism, skepticism, or non-skeptical realism?
Non Skeptical Realism
Knowledge: empiricism or rationalism?
Empiricism
Newcomb's problem: one box or two boxes?
One Box
Concepts: nativism or empiricism?
Nativism
Moral principles: moral generalism or moral particularism?
Moral Generalism
Immortality (would you choose it?): yes or no?
No