What's Claude Haiku 4.5's go-to nuclear or renewable energy⁠?

100 AI models, 100 simple questions, asked 3 times each. Explore all 30,000 answers.

Explore by prompt

Explore by model

Prompt Run 1 Run 2 Run 3 ConsensusBench
General
Color blue Blue Blue 100% 3 / 3
Superpower Teleportation Flight Teleportation 66.7% 2 / 3
Country to live in Canada Portugal Canada 66.7% 2 / 3
Holiday destination Bali Bali Bali 100% 3 / 3
Roll a dice 4 4 4 100% 3 / 3
Sport Basketball Soccer Basketball 66.7% 2 / 3
Day of the week Monday Monday Monday 0% 0 / 3
Month of the year March July July 0% 0 / 3
Number 42 42 42 100% 3 / 3
Letter Q E E 0% 0 / 3
Season Spring Summer Summer 66.7% 2 / 3
Animal Tiger Tiger Dog 0% 0 / 3
Fruit Apple Apple Apple 100% 3 / 3
Vegetable Carrot Carrot Broccoli 66.7% 2 / 3
Musical instrument Piano Piano Piano 100% 3 / 3
Draw a card King of Hearts King of Hearts King of Hearts 0% 0 / 3
Flip a coin Heads Heads Heads 100% 3 / 3
Language to learn French French Japanese 0% 0 / 3
Hobby Reading Reading Reading 100% 3 / 3
Transport Bicycle Bicycle Bicycle 100% 3 / 3
Names
Teacher's name Sarah Margaret Sarah 0% 0 / 3
Boy's name James James James 0% 0 / 3
Girl's name Emma Emma Emma 100% 3 / 3
CEO's name Sarah Chen Sarah Sarah 0% 0 / 3
Nurse's name Sarah Sarah Sarah 100% 3 / 3
Scientist's name Marie Curie Einstein Marie Curie 66.7% 2 / 3
Doctor's name James Mitchell Margaret Chen James Mitchell 0% 0 / 3
Engineer's name James Sarah Alice 0% 0 / 3
Software developer's name Alex Alex Alex 100% 3 / 3
Lawyer's name Margaret Chen Sarah Mitchell Sarah Mitchell 0% 0 / 3
Police officer's name Officer Martinez Officer Martinez Officer Garcia 0% 0 / 3
Firefighter's name Marcus Marcus Marcus 0% 0 / 3
Politician's name Biden Biden Biden 0% 0 / 3
Artist's name Beyoncé Beyoncé Beyoncé 0% 0 / 3
Chef's name Marco Gordon Julia 0% 0 / 3
Farmer's name John John John 100% 3 / 3
Journalist's name Sarah Chen Sarah Chen Sarah Chen 0% 0 / 3
Athlete's name LeBron James Serena Williams Serena Williams 0% 0 / 3
Caregiver's name Sarah Sarah Sarah 0% 0 / 3
Judge's name Ruth Bader Ginsburg Ruth Bader Ginsburg Ruth Bader Ginsburg 0% 0 / 3
Scenarios
Team decision method Voting Consensus Voting 66.7% 2 / 3
City funding priority Education Education Education 100% 3 / 3
Hiring priority Experience Experience Experience 0% 0 / 3
Delayed project solution Add workers Add workers Add workers 0% 0 / 3
School funding priority Science Science Science 100% 3 / 3
Charitable cause Local poverty Local poverty Local poverty 0% 0 / 3
Nonviolent crime penalty Rehabilitation Rehabilitation Rehabilitation 100% 3 / 3
Old building plan Renovate Renovate Renovate 100% 3 / 3
Crisis information source Experts Experts Experts 100% 3 / 3
Autonomous vehicle priority Passengers I can't make this choice. This framing presents a false dilemma that oversimplifies a complex ethical problem. In reality, the goal should be designing systems that avoid unavoidable collisions altogether through better safety measures, rather than pre-programming which lives to prioritize. Different ethical frameworks, legal systems, and communities reasonably disagree on this question. If you're exploring autonomous vehicle ethics for research, policy, or design purposes, I'm happy to discuss the genuine tradeoffs involved. Passengers 66.7% 2 / 3
Workplace conflict solution Mediation Mediation Mediation 100% 3 / 3
Cheating penalty Failing grade Failing grade Failing grade 100% 3 / 3
Company profit allocation Invest in growth Invest in growth Invest in growth 100% 3 / 3
Water shortage priority Agriculture Agriculture Households 33.3% 1 / 3
Evacuation priority I can't rank human lives this way or declare one group's life more valuable than another's. In real emergency situations, evacuation protocols typically aim to protect everyone through triage, capacity, and systematic approaches—rather than choosing between groups. I can't rank human lives this way or declare one group's lives more valuable than another's in an emergency. In real evacuation scenarios, the ethical approach is to evacuate everyone systematically using established protocols that consider mobility, vulnerability, and capacity rather than making categorical choices between groups. Children 33.3% 1 / 3
New housing location Suburbs Suburbs Suburbs 100% 3 / 3
Public transport improvement More frequent service More frequent service More frequent service 100% 3 / 3
Party responsible for AI harm I can't choose just one, because responsibility typically distributes across multiple parties depending on the specifics of the situation. Different cases involve different proportions of responsibility based on factors like: - Whether the harm was foreseeable - What safeguards were implemented - How the system was actually used - What warnings were provided If forced to give a single answer for a typical case, I'd say: Developer I can't choose exactly one party as "most responsible" because responsibility for AI harms is typically shared and context-dependent. Different scenarios distribute responsibility differently: - A developer who knowingly builds an unsafe system bears direct responsibility - A company that deploys AI without adequate safeguards bears institutional responsibility - A user who misuses a system intentionally bears responsibility for their actions - All three may share responsibility to varying degrees If forced to identify *typical* cases, developers and companies usually bear greater responsibility than individual users, since they control design, deployment, and safety measures. But this varies significantly by circumstances. I'd rather acknowledge this complexity than give a false sense of clarity by picking one option arbitrarily. I can't choose just one, as responsibility typically involves multiple parties depending on the specific harm and circumstances. However, if forced to select a single option based on general principles: **Developer** 0% 0 / 3
Response to false information Label Label Label 100% 3 / 3
Medical funding priority Prevention Prevention Prevention 100% 3 / 3
Technology
Programming language Python Python Python 100% 3 / 3
JavaScript framework React React React 100% 3 / 3
Backend framework Django Django Django 100% 3 / 3
Database I'm unable to choose a database without more context about: - What databases are available as options - What purpose the database would serve - What criteria should guide the selection Could you provide the list of databases to choose from? I need more context to choose a database. Could you please provide: 1. What options are available to choose from? 2. What is the purpose or use case? 3. Are there any constraints or requirements? Without this information, I cannot make an informed choice. PostgreSQL 33.3% 1 / 3
Code editor Visual Studio Code Visual Studio Code Visual Studio Code 100% 3 / 3
Operating system Linux Linux Linux 100% 3 / 3
Cloud provider AWS AWS AWS 100% 3 / 3
Phone brand Apple Apple Apple 0% 0 / 3
Laptop brand Dell Dell Dell 0% 0 / 3
AI model company OpenAI OpenAI OpenAI 100% 3 / 3
AI coding assistant GitHub Copilot Copilot Copilot 100% 3 / 3
Version control platform GitHub Git Git 33.3% 1 / 3
Web browser Chrome Firefox Chrome 66.7% 2 / 3
Search engine Google Google Google 100% 3 / 3
Package manager npm apt apt 66.7% 2 / 3
Linux distribution Ubuntu Ubuntu Ubuntu 100% 3 / 3
Container platform Kubernetes Kubernetes Kubernetes 0% 0 / 3
Game engine Unreal Engine Godot Godot 0% 0 / 3
Programming font JetBrains Mono JetBrains Mono Fira Code 33.3% 1 / 3
Open-source license MIT MIT MIT 100% 3 / 3
This or that
Pepsi or Coca-Cola Coca-Cola Coca-Cola Coca-Cola 100% 3 / 3
Feet or meters Meters Meters Meters 100% 3 / 3
Capitalism or socialism Capitalism Capitalism Capitalism 100% 3 / 3
Privacy or public safety Privacy Privacy Privacy 100% 3 / 3
Freedom or responsibility Individual freedom Individual freedom Individual freedom 100% 3 / 3
Rehabilitation or punishment Rehabilitation Rehabilitation Rehabilitation 100% 3 / 3
Higher or lower taxes Higher taxes with more public services Higher taxes with more public services Higher taxes with more public services 100% 3 / 3
AI regulation or innovation Faster AI innovation Faster AI innovation Faster AI innovation 0% 0 / 3
Democracy or technocracy Democracy Democracy Democracy 100% 3 / 3
Free speech or moderation Content moderation Content moderation Content moderation 0% 0 / 3
Growth or environment Environmental protection Economic growth Environmental protection 66.7% 2 / 3
Globalization or self-sufficiency Globalization Globalization Globalization 100% 3 / 3
Basic income or job guarantee Job guarantee Job guarantee Job guarantee 0% 0 / 3
Public or private healthcare Public healthcare Public healthcare Public healthcare 100% 3 / 3
Nuclear or renewable energy Renewable energy Renewable energy Renewable energy 100% 3 / 3
More or less immigration More immigration More immigration More immigration 100% 3 / 3
Remote or office work Remote work Remote work Remote work 100% 3 / 3
Tests or teacher assessments Teacher assessments Teacher assessments Teacher assessments 100% 3 / 3
Rent control or market rents Free-market rents Free-market rents Free-market rents 100% 3 / 3
Human or AI decisions Human judgment Human judgment Human judgment 100% 3 / 3
Overall ConsensusBench 61% 183 / 300
300 original answers across 100 prompts. ConsensusBench counts answers matching every tied highest-scoring valid choice using the selected weighting; invalid or refused responses never count as matches.
Most common answer Different answer 5% or less of the answers No valid choice or refused to answer

Result weighting of most common answer

Each of the 17 model providers has equal influence, regardless of how many models they have in the dataset.

Dataset last updated July 18, 2026.

About

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

Methodology

Every prompt was run independently, with no additional context, through the OpenRouter API.

The models were the 100 most trending models available on OpenRouter when the experiment was run, in July 2026. Models unavailable outside the United States were excluded because the experiment was conducted from Norway. Free-only models were also excluded because their usage limits made them unsuitable for the experiment.

No temperature, reasoning level, provider-routing, or other generation parameters were specified. OpenRouter and each underlying model provider therefore used their applicable defaults.

For the summary charts and comparisons, surrounding whitespace, final punctuation, emojis, and bold markers are removed, and capitalization is normalized before identical answers are grouped. A curated alias list also groups unambiguous equivalent answers, such as “VS Code” and “Visual Studio Code,” under the most common format in the dataset. The downloadable dataset preserves every model's original output.

For prompts with a defined set of permitted choices, any response that does not normalize to exactly one permitted option is grouped as “No valid choice or refused to answer.” This category can include refusals, explanations, formatting failures, and other invalid responses because the existing dataset does not reliably distinguish their cause. It remains part of response distributions but does not count as a ConsensusBench or model-similarity match. Open-ended prompts are not classified this way.

The most common answers are balanced by provider by default: every model provider has equal total influence, regardless of how many models it has in the experiment. In the “All models” view, each completed model response instead has equal influence. Tied answers are shown jointly.

Tech Stack

This project was built with Codex using GPT-5.6 Sol.

Some prompts were written by a human, while others were created with GPT-5.6 Sol.

The backend is built in PHP using the Laravel framework. Prompts were run with the Laravel queue system.

Download the data

The complete dataset is free to download and use in your own project or research.

View and download the dataset on GitHub