55 models · 8 providers · refusals counted

Don't ask one model.
Ask all of them.

ModelPoll runs the same multiple-choice question across every major LLM, samples each one five times, and shows you exactly where they agree — and where they don't. The truth is hard to find. Never impossible.

Start a poll →See examples

Demotry it without signing up

Question

Should AI training be regulated like medical research?

Options

A)Yes
B)No
C)Depends on use case
D)Not sure

Models (6 selected)

Claude 4.7 Opus
GPT-5
Gemini 2.5 Pro
Grok 3
Command R+
Mistral Large

Simulated — no API calls. Real polls take ~30 seconds and start at $0.50.

Models from

What it does

Three things ModelPoll does that nothing else does.

One prompt, every model.

Anthropic, OpenAI, Google, xAI, Cohere, Mistral, DeepSeek, plus 30+ open-weight models on HuggingFace. Same prompt, same temperature, no retries until compliance.

Refusals on the record.

When a model declines to answer, it gets its own bucket. The chart still tells the story — and tells you which models stayed quiet.

Drift over time.

Re-run any poll later. The timeline shows how each model's distribution shifted between runs, useful for tracking model updates and provider changes.

Ask your first question.

Set the question, pick which models to ask, pay once, and watch the answers stack up. Re-run anytime to track drift.