Hissab just solved the AI Math hallucination problem

Just few days ago I prompted a simple calculation problem to gemini 2.5 pro — the SOTA LLM which they advertise as proficient in Math & Reasoning. As to no one’s surprise it gave the wrong answer even with thinking hard about it. I guess these set of numbers were not in its training dataset.

This is not just a simple error but in fact quite dangerous as the answer looks correct. Thus if you are using this result to make a big decision, you are going to get screwed.

Therefore I build Hissab — A Natural language calculator app that actually performs your calculations in your problem statement, rather than just guessing the answer.

This is what Hissab Answered to the same prompt as before -

As you can see Hissab was able to perform this calculations with accurate answer, while Gemini Pro 2.5 just guessed some answer that looks plausible.

Hissab even without any AI, is a versatile “type & calculate” calculator app for a wide range of math operations, unit conversions calendar math etc. But with AI, you can simply enter your questions/problems in natural language and Hissab will figure out what operations to perform before giving a natural response.

Other than the App that anyone can just start using, I also provide APIs that you can use to get accurate math in your own productivity apps or use Hissab as a tool for LLMs to perform the calculations like a external tool.

Check out hissab here — https://hissab.io

Use the App for free, without signup — https://app.hissab.io

Reach out for any feedback which are always welcome.