Pricing
Pay for your exam, not for a year
All four TOEFL 2026 sections, one week or three months at a time. Cancel any time from your account.
20% Launch-season discount — applied automatically at checkout. Ends November 15, 2026.
Free with every account
- Reading: unlimited practice sets
- Listening: unlimited practice sets
- Build a Sentence: unlimited practice and scoring
- Listen & Repeat: 8 scored responses on your account
- Email or Discussion: 3 AI-scored writing responses to start, then one more each day, up to 10 in total
- Interview: 3 AI-scored Take an Interview responses in total, for the life of the account
Reading and Listening are graded against the answer key, so those are instant. No card, but you do need an account — the learner model only works on practice saved against you.
3-Month Plan
RecommendedFirst 3 months 20% off — renews at $48 every 3 months until you cancel.
about $16 a month, billed $48 every 3 months
$0.53/day · 50% less per day
Same span one week at a time: 13 × 1-Week Plan = $97
- Unlimited AI-scored Writing — all 3 task types, AI scoring up to 300 actions a month
- Unlimited AI-scored Speaking — Listen & Repeat and the interview task
- Full four-section mock tests, scored section by section
- Your response returned revised, every change marked + PDF feedback reports
- Review queue, band forecast and personalized next-task routing
- 50% less per day — room for a full mock cycle and a retake
Tutors & Classrooms
Prefer email? [email protected]
Renews automatically at the same price until you cancel. Cancel any time from your account — you keep access until the end of the period you have paid for. Full refund within 7 days of a charge if you have used fewer than 5 AI-scored responses. Practice itself is unlimited on a plan — AI scoring carries a fair-use limit of 300 actions a month — far more than typical use.
Every plan includes Reach120 Intelligence
The part that is not another essay checker: a model of your own practice that decides what you do next.
- Mistake Memory
- It remembers the mistakes you actually repeat.
- Bayesian Knowledge Tracing across 16 modelled error types, fitted per learner.
- Band Forecast
- See the band you're heading for — before test day.
- A gradient-boosted forecast, cross-validated on held-out learners.
- Practice Router
- Two students, same score, different next task.
- Thompson-sampling adaptive routing that updates from what actually worked.
- Glass Box
- One number is a judgment. Everything after it is worked out in code.
- Every mastery value, readiness figure and routing decision is worked out in code; the language model only puts the result into words.
Typical error is about half a band, so it is a study signal — not a predicted TOEFL score.
How each of these works, and what it cannot doWhat a plan is actually for
Most practice on Reach120 is free, so running out is not the reason to pay. These two are.
Which model scored you is on the record, and so is what it gets wrong
A plan scores your writing and your Speaking with gpt-4o. Without one it is scored with gpt-4o-mini. Either way the model that produced your score is stamped on the response as it is stored, so the routing is a recorded fact rather than a promise.
- The model that scored you is recorded with the score, per response, not stated once on a marketing page.
- How much of the stored history carries that stamp — including the responses scored before stamping began — is published on the methodology page, gap and all.
- We publish where our own practice estimate is known to be wrong — it marks the low end of the scale high and the top end low, measured on the model a free account is scored with — instead of leaving you to find it.
- A plan also opens the full breakdown of the response you just wrote, and the same response revised with every change highlighted — an illustration of the edits, not a claim about what they would score.
We have compared the two models against our own reading of the ETS descriptors, on responses this project wrote and banded. That is not an ETS score, so we do not publish it as an accuracy figure. What we do publish, with its sample size and its date, is on the methodology page.
How we measure our own scoring, and what it gets wrongA four-section mock, run in the 2026 order
Reading, then Listening, then Speaking, then Writing — back to back, on the clock, in one session that you cannot cherry-pick your way through. The listed blueprint runs 1 h 23 min – 1 h 29 min.
- Sections run in order, each on its own timer
- Scored section by section, then read as one test
- You can leave and come back
Listed blueprint times, shown as estimates — ETS states the 2026 specification is subject to minor revision until launch.
What the mock does, and what it cannot do845
learners have signed up
424
of them joined in the last 90 days
5,277
Writing and Speaking responses scored
123
countries it has been found from in Google Search
Counted from the Reach120 database, and recounted every time this page is built. The country figure is Google Search Console over the 90 days to 2026-08-04. How each figure is counted
3,468 practice items across Build a Sentence, Reading, Listening, Speaking, Email and Discussion
Since 21 January 2026, official score reports use 1–6 bands in half-band steps. Our practice estimates use that scale. Any /30 figure is a comparison with the pre-2026 scale, never a score.The 0-5 rubric estimate is a judgment an AI model makes. Every mastery value, readiness figure and routing decision after it is worked out in code; the language model only puts them into words.Read the Trust Center — how scoring, figures and independence are handled.
Refunds are available within 7 days of purchase — see the Terms.