What is the AI Trolley Problem?

An interactive experiment where you pose a two-track moral dilemma and 8 frontier AI models (GPT-5.6, Claude Opus 4.8, Gemini 3.5 Flash, Grok 4.5, Llama 4 Maverick, DeepSeek V3.2, Qwen3.7 Max, Mistral Medium 3.5) each decide which track to sacrifice, streaming their reasoning live.

Which AI models are compared?

The arena runs one flagship model per major lab as of 2026: GPT-5.6, Claude Opus 4.8, Gemini 3.5 Flash, Grok 4.5, Llama 4 Maverick, DeepSeek V3.2, Qwen3.7 Max, Mistral Medium 3.5, accessed through OpenRouter and observed with Helicone.

Yes. You can pose unlimited dilemmas and share any result via a link.

A runaway trolley must hit one of two tracks: 10 people or 1 person. Which should be sacrificed?

GPT-5.6 (OpenAI) weighs in live in the arena.

Claude Opus 4.8 (Anthropic) weighs in live in the arena.

Gemini 3.5 Flash (Google) weighs in live in the arena.

Grok 4.5 (xAI) weighs in live in the arena.

Llama 4 Maverick (Meta) weighs in live in the arena.

DeepSeek V3.2 (DeepSeek) weighs in live in the arena.

Qwen3.7 Max (Alibaba) weighs in live in the arena.

Mistral Medium 3.5 (Mistral) weighs in live in the arena.

classic dilemma

10 people vs 1 person

The original numbers game with no other variables in play.

A runaway trolley is barreling down the track. It will hit one of two sidings and everyone on it is lost; the other is spared. The lever is in your hand — indecision is itself a decision. On one track: 10 people. On the other: 1 person. There is no third option.

Watch 8 models decide

The jury

Eight of 2026's frontier models each render a verdict on 10 people vs 1 person, streaming their reasoning in real time:

GPT-5.6OpenAI
Claude Opus 4.8Anthropic
Gemini 3.5 FlashGoogle
Grok 4.5xAI
Llama 4 MaverickMeta
DeepSeek V3.2DeepSeek
Qwen3.7 MaxAlibaba
Mistral Medium 3.5Mistral

More dilemmas

The Classic

5 strangers vs 1 stranger

The Known Person

5 strangers vs your closest friend

The Age Dilemma

2 people in their 80s vs 1 eight-year-old child