A tool you can trust is one whose limits you know. These are Keplar's, as of 3 October 2026.
It can be wrong, including when the models agree
Models trained on overlapping data can share a mistake. Keplar's agreement readout counts how many models took the same position. It does not measure correctness, and Keplar shows no confidence score or accuracy percentage because it has none to show. See Why consensus can be wrong.
It does not browse the web today
The Web option in the composer is labeled as not live. Models answer from what they learned in training, so anything that changes weekly (current software versions, prices, release dates, news, sports results) can be out of date. In a test on 3 October 2026, Keplar's free panel gave out-of-date answers on a "current LTS version" question; it flagged that the models split but could not look up the right answer. Check recent facts at the official source.
It does not check the sources models cite
The Sources section lists links the models gave. Keplar has not opened them. A model can cite a page that does not say what it claims, or a page that does not exist. Open any link you rely on.
Simple questions use one model
A short, clear question goes to one fast model and is not cross-checked, unless it involves counting, arithmetic or an exact figure, which always get a small panel. A single model can still slip on a simple question. If it matters, rephrase it or choose Most thorough, which gives even simple questions a panel.
It cannot read your code base, your inbox or your files by itself
Keplar reads what you attach to a question and what you connect on purpose, and only the file types the attach button accepts today. Scanned PDFs cannot be read because there is no text recognition (OCR) yet.
It is not professional advice
Answers on health, law and money are general information. Where a decision matters, talk to a qualified professional.
It does not make models cheaper or smarter
Combining models can surface errors and trade-offs a single model hides. It can also cost more than a single call. Keplar is a way to compare and synthesize, not a guarantee of a better answer on every question.
Free-plan caveats
Free answers can be slower, and the hosts of free models may log prompts. Do not put sensitive information in a Free question.
If you find an error
Tell us at team@keplar.one or through the feedback form. Specific examples help most.
Related
- Why consensus can be wrong: The limits of agreement between models: shared training data, shared blind spots, stale knowledge and leading questions, and what to do about each.
- Sources, citations and web lookups: What the Sources section really contains, why Keplar does not open or check the links, the status of web lookups, and a checklist for verifying what matters.
- How Keplar handles your data: Where a question goes, what is stored under your account, which companies process it, what is never sent to model providers, and what Keplar does not do.