Simple, usage-based pricing
Start free. Scale as you grow. No commitments, no hidden fees.
Frequently asked
What counts as one API call?
Each POST to /v1/score is one call. If you include multiple statements in one request, each counts as one call.
Can I use my own LLM?
Yes. ConfiScore works with any LLM — GPT-4, Claude, Llama, Mistral, or your own fine-tuned model. The score is based on the text and context, not the model.
What are Platt scaling parameters?
If you've calibrated your model, providing the Platt scaling coefficients (a, b) improves ConfiScore's accuracy. Not required — the API works without them.
How is this different from self-consistency?
Self-consistency requires 5–20 inference calls per question. ConfiScore needs one. We achieve lower Expected Calibration Error (ECE) at a fraction of the cost.
Can I upgrade or downgrade anytime?
Yes. All plans are month-to-month. Change anytime. Unused calls don't roll over.