A free digital SAT practice tool that analyses how you think — it reads your highlights, eliminations and timings to show why you got a question wrong, not just that you did.
Total 1200
All of this lives in your Profile and can be changed later.
Digital SAT - practice under real conditions
Find your level in about twenty minutes a section. Every question is chosen from how you answered the last one, so the test walks to the hardest material you can actually handle and estimates your score from where it settles.
How long you take is read too. An answer given fluently counts for more than one ground out, and a snap wrong answer counts for less than a considered miss.
Drill one skill at a time. This is not a timed mock - pick what you want to work on and how much of it.
Have an AI write you a fresh set. Choose what you want, pick your model, and the prompt is copied ready to paste.
A lesson rather than a paper. Open your AI in voice mode with Colorbook connected and it sets the questions one at a time — watching what you pick, what you rule out and how long you take, and marking up the passage in its own colour while you talk it through.
Start the session here first — your tutor cannot reach a screen that is not already listening. Then tell it what you want to work on.
This tool is for rehearsing the exam, not for meeting the material the first time. When a topic is genuinely new, these are the places to go and learn it — then come back and sit a paper.
Two modules, 32 minutes each.
Keep talking - the next question will appear here.
Return to the section to review or complete any outstanding answers.
You may continue to the next module once the timer reaches zero.
Do not close this device.
Open any question below and you get the paper back exactly as you left it - your highlights, your notes, what you crossed out - with a timeline of everything you did and when you did it. A score tells you how many you got. This tells you why.
Every question with your answer, what you eliminated, and the explanation.
Read the trap column before the topic column - several misses sharing a trap are one habit, not several weaknesses.
What you put, what was right, and why.
A Markdown file with every question, your answers, the full annotation timeline, and instructions telling an AI exactly how to analyse it. The one-press hand-off to a model is on the Summary tab.
Your whole history across every attempt on this device - every bank question you've answered and whether you got it right. If you clear your browser data or move machine, paste it into "Import done questions" to pick up where you left off.
You left the test window. On test day this ends your session. Return to fullscreen to continue.
Add any official scores you already have. They anchor the graphs on the Analytics page.
A portable copy of everything you've answered. Save it, and you can restore your history on another device.
With no timer the report says "untimed" and skips pacing analysis, since minutes per module stop being meaningful.
Text-to-Speech uses your device's built-in voice. A speaker control appears on each question while it is on.
The digest is your section scores, topic tallies, how you worked, and each missed question shortened to its stem plus the choice you put and the one that was right. Passages and your name are never sent. Turn this off and nothing is transmitted at all - the rest of the results page works exactly the same without it. Limited to one summary every six hours.
Give an AI model a direct line to your question bank and your own attempts, instead of pasting one question at a time. It can look a question up by code, read the explanation, and see how you actually worked a paper.
Two ways, depending on what your client asks for. If it has a place for headers, use Copy config. If it only asks for a URL, use Copy connect URL — the token travels in the address, so it authenticates on its own with nothing else to fill in.
Every tool is read-only. A connected model can read the question bank and your own attempts; nothing it calls can change an answer, edit the bank, or touch your account. It only ever sees your data — the token is what identifies whose attempts to serve.
Treat the token like a password: anyone holding it can read your attempts. Signing out replaces it, which disconnects any client still using it.
The connect URL is the easier of the two and the less private: a token in an address can end up in server logs and in whatever history your client keeps, where a header usually does not. Prefer the config form when your client accepts one.
Working it out…
Sat a real SAT? Upload the score report PDF and your weakest content areas come straight from it.
Colorbook is in active development. You may run into bugs, a question with the wrong answer or a broken figure, or a feature that changes from one visit to the next. Your practice history is kept, but please do not rely on a score here for anything that matters.
If something looks wrong, telling us is the fastest way to get it fixed.