What is self-consistency / majority voting?

Self-consistency is a prompting technique where you sample the same prompt several times at non-zero temperature and keep the answer the model produces most often. It typically improves accuracy on reasoning tasks versus a single sample.

How are answers grouped into votes?

Each response is normalized — trimmed, lowercased and stripped of surrounding punctuation and whitespace — then exact-matched. Responses that normalize to the same string share a vote cluster, so trivial formatting differences do not split the vote.

Does this cost more than one call?

Yes. Running N samples makes N separate API calls and bills N times the tokens of a single request. The tool runs them in parallel against your own key, so you pay your provider directly.

Your key stays in the browser tab and is sent only to the provider's official endpoint in each request. It is never logged, stored or transmitted to Gera.

Multi-Output Majority Voter (BYO-key)

Run a prompt N times and keep the majority answer

Large language models are non-deterministic: ask the same question twice at a normal temperature and you can get two different answers. Self-consistency turns that variance into a reliability signal. This tool sends your prompt to your own OpenAI or Anthropic key N times, groups the responses by content and surfaces the majority answer along with how strongly the runs agreed.

How it works

Each of the N requests is an independent API call at the model’s default sampling. When all responses return, the tool normalizes every answer — trimming whitespace, lowercasing and stripping wrapping punctuation — and then buckets identical normalized strings together. The bucket with the most votes is the majority answer, and the agreement percentage (winning votes ÷ N) tells you whether the model is confident (one dominant cluster) or unstable (many small clusters). High disagreement is itself useful: it means the task is ambiguous, the prompt is under-specified, or the answer is genuinely uncertain.

Tips and notes

For short, factual or classification-style answers, majority voting works best — normalize the task so the model returns a single token or short phrase (“yes/no”, a label, a number). For long free-form generations exact-match clustering will rarely agree, so use this on the final extracted answer rather than full prose. Odd values of N avoid ties. Five to seven runs is a practical balance of cost and stability; push to nine or more only for genuinely high-stakes decisions. Remember every run is billed separately against your key.