It builds a model of you, and routes your practice from it

Reading, Listening and Build a Sentence are free for every account — checked against a stored answer key, so there's no daily or lifetime limit. Listen & Repeat gets 8 free scored responses on the account. Write an Email, Academic Discussion and Take an Interview are scored by AI, which costs real money per attempt, so they are metered in two separate lanes: Write an Email and Academic Discussion share 3 AI-scored writing responses to start, then one more each day, up to 10 in total, and Take an Interview has 3 AI-scored Take an Interview responses in total, for the life of the account — and a plan lifts both.

There is no separate trial, because the free account is not a preview — it includes unlimited Reading, Listening and Build a Sentence practice, 8 scored Listen & Repeat responses on the account, and 3 AI-scored writing responses to start, then one more each day, up to 10 in total across Write an Email and Academic Discussion, plus 3 AI-scored Take an Interview responses in total, for the life of the account, each with the full report, plus the sample essays and the topic library. What a plan adds is AI-scored Writing and Speaking with no allowance to run out, and the full mock tests: $7.49 every week or $48 every three months, cancel any time.

From your own record, not a fixed syllabus. Each scored response updates a mastery estimate across 16 modelled error types, and the next task is chosen against those estimates — which is why two people with the same score can get different next tasks. In Writing and Listening that choice runs through adaptive routing that updates as learners use it; Reading and Speaking use per-learner rules over the same record.

No. Every score, mastery value and recommendation is worked out by fixed rules in code, not by the language model. The model only puts those results into words, and a written explanation containing a number or a pattern the rules did not produce is rejected rather than shown to you. Reading and Listening are graded against the answer key, so those results are exact.

It estimates the band your next practice session is heading for, section by section. Typical error is about half a band, so it is a study signal — not a predicted TOEFL score. It stays blank until you have practiced enough for the estimate to mean anything, rather than showing you a confident-looking number built on two sessions.

AI does. The rubrics and feedback criteria are Reach120’s own, written against the 2026 TOEFL Writing and Speaking task descriptions ETS publishes — they are not ETS rubrics, and Reach120 is independent of ETS. Build a Sentence, Reading and Listening are checked against a stored answer key instead, so those results are exact.

All four TOEFL 2026 sections: Reading, Listening, Speaking, and Writing. Writing covers the three 2026 task types (effective January 21, 2026) — Build a Sentence, Write an Email, and Academic Discussion — and it’s the section we go deepest on, since it’s the hardest to grade yourself.

Reading, Listening and Build a Sentence are uncapped on any account — no daily limit and no lifetime limit. Listen & Repeat comes with 8 free scored responses on the account, which do not refill. Write an Email, Academic Discussion and Take an Interview are AI-scored, so a free account gets 3 AI-scored writing responses to start, then one more each day, up to 10 in total across Email and Discussion, and 3 AI-scored Take an Interview responses in total, for the life of the account for Interview. Only the daily Writing unit resets each morning — the units you start with, and Interview's, do not come back once spent. A subscription removes both limits and practice is unlimited. AI scoring carries a fair-use limit of 300 actions a month — far more than typical use.

Yes — a plan renews automatically at the same price, every one week or three months, until you cancel. We email you before each renewal. You can cancel in two clicks from your account page: that stops all future charges and you keep full access until the end of the period you have already paid for. A charge is refundable within 7 days if you have used fewer than 5 AI-scored responses in the period it paid for.

No. Official scores are issued only by ETS, and Reach120 shows a practice estimate. Its criteria reference the publicly available task and rubric descriptions for the 2026 format, but how closely the estimate matches official scoring has not been measured, so we do not claim it does. The useful part is the reasoning and the line-level feedback, which you can check against your own writing.

Partly, and it is worth being exact about which part. The tasks, the timers and the rubrics are TOEFL-shaped, so that half is exam-specific. The error record is not: the feedback names which of the 16 modelled error types a sentence hit, and returns your response revised with every change marked, so you see the correction rather than a grade. Article use, tense agreement and sentence boundaries do not stop mattering the day after the test. Whether that carries into your work email is not something we have measured, so we do not put a number on it.

There is no score guarantee here, because ETS issues the score and we do not — anyone offering one is selling you something they cannot deliver. What this does instead is make the gap legible: the feedback names the issues in each response, the progress view shows whether those specific issues are moving, and the band estimate updates section by section as you practice. On the money: a charge is refundable within 7 days if you have used fewer than 5 AI-scored responses in the period it paid for. That window is about unused practice, not about your result — a refund tied to your score would be a claim about your score, and we do not make one.

Your responses are stored because the product is built on them — the mastery estimates, the progress view and the choice of your next task all read your own record, and there would be nothing to read if we discarded it. Selected practice responses may appear in the public samples section with personal names removed. We do not sell your data: it goes to the services that run the product (Supabase for authentication, Stripe for payments) and to analytics in anonymised form. Deleting your account from Account Settings removes your submissions, feedback, ratings and saved payment methods immediately rather than on a schedule — the one exception is the invoices Stripe keeps for its own accounting. The Privacy Policy has the full list.