Think hard about the words you send.

Why it exists

Teams change prompts every day and judge the result from a couple of answers on one model. The tools that do this properly, with test sets, versions and evaluations, are built for engineering teams with an SDK, an account and a budget. Most people who write prompts have none of those.

thinkhard.si puts the same discipline in a browser tab: grade a prompt against ten plain checks, run it against real inputs on the models you are choosing between, price it at your volume, and attack it before someone else does.

Principles

• Useful before anything is asked of you. Every page does real work without a sign-up, and most without a key.
• Your work is yours. Prompts, keys and results stay in your browser. Model calls go straight to the provider you pick.
• No fabrication. Example data is labeled as an example wherever it appears. Estimates say they are estimates.
• Plain words. Hardness is a count of tests passed, not a mood. Prices carry the date they were checked.
• Free stays free. Paid plans will only cover what needs a server.

The name

Read the address out loud: think hard, SI. SI is short for super intelligence, the name the US government has used for AI in its official documents since September 2026. “Think hard” is the instruction every good prompt gives it, and a careful answer starts with a carefully written request.

The mark is a speech bubble cut like an emerald: what you say to the model, shaped with care, with SI in gold at its heart. The minerals around the site are a nod to the other kind of hard, from talc, which crumbles under a fingernail, to diamond, which nothing natural scratches.

Get in touch

thinkhard.si is a small, independent project, and the person who builds it reads every message. Found a bug, a wrong price, or a tool you wish existed? Email contact@thinkhard.si or use the contact form.

Changes are listed in the changelog.