Skip to main content
πŸ“– The AI Tool Bible
ModelBias preview image
ModelBias logo

ModelBias

βœ“ Text source-checked Nov 22, 2026

100 AI models, 100 simple questions, asked 3 times each, with 30,000 answers to explore.

Source-checked against the vendor's site Β· rebuilt from the vendor's own pagesProfile updated Link checked How we rate

FreeEvaluation100 models across Anthropic, OpenAI, Google, DeepSeek, Meta, xAI, Mistral, Qwen and others (via OpenRouter) (not source-checked)

ModelBias.ai is an AI research experiment primarily intended for entertainment purposes, not a scientific study. The vendor says that the project attempts to highlight the default choices and biases models can exhibit when no additional context is provided.

The vendor says that every prompt was run independently, with no additional context, through the OpenRouter API. The models tested were the 100 most trending models available on OpenRouter when the experiment was run in July 2026. The vendor says that no temperature, reasoning level, provider-routing, or other generation parameters were specified for the runs.

The vendor says that the most common answers are balanced by provider by default, giving every model provider equal total influence regardless of how many models it has in the experiment.

The complete dataset is free to download and use in one's own project or research. The dataset is available to view and download on GitHub.

From the vendor's pages

According to ModelBias: ModelBias.ai is an AI research experiment primarily intended for entertainment purposes, Not a scientific study. Its pages list this strength: The complete dataset is free to download and use in one's own project or research. They also state this limitation: Models unavailable outside the United States were excluded from the experiment.

β€” The AI Tool Bible editorial team

Pros

  • βœ… The complete dataset is free to download and use in one's own project or research.
  • βœ… The vendor says that aggregates are balanced by provider by default, ensuring every model provider has equal total influence.
  • βœ… The vendor says that the downloadable dataset preserves every model's original output.
  • βœ… Users can explore results by choosing a specific prompt.
  • βœ… Users can explore results by choosing a specific model.

Cons

  • ⚠️ Models unavailable outside the United States were excluded from the experiment.
  • ⚠️ Free-only models were excluded because their usage limits made them unsuitable for the experiment.

Facts

Open source
Yes (not source-checked)
Model used
100 models across Anthropic, OpenAI, Google, DeepSeek, Meta, xAI, Mistral, Qwen and others (via OpenRouter) (not source-checked)
Hacker News
1 story Β· top 8 points Β· last Jul 23, 2026 Β· checked Nov 16, 2026
Earliest Wayback capture
2026-07-26 (earliest archived capture of the site, not a launch date)

Where this tool sits

Use cases

Comparing hidden defaults, associations, and biases of AI modelsUsing the dataset in one's own project or researchHighlighting default choices and biases models exhibit without additional context
See alternatives to ModelBias β†’

Explore related

Reviews