The official TOEFL 2026 test blueprint, in full
Every figure below is transcribed directly from ETS's own test blueprint — the 120 items, 145 raw points, per-task item counts, CEFR targets, and stimulus specifications aren't inferred from practice tests or estimated from experience. They're the exact numbers ETS publishes for this purpose, not a paraphrase of them.
The full task-by-task blueprint
Every task in the exam, grouped by section, with ETS’s own name for it, its item count, the CEFR level it targets, and whether it’s machine- or AI-scored.
| ETS task name | Items | CEFR target | Scored by | Our guide |
|---|---|---|---|---|
| Reading · 50 items · 35 raw points · adaptive — router 18–21 min, module 9 min | ||||
| Complete the Words | 30 | B1–C1+ | Machine | Complete the Words |
| Read in Daily Life | 5–15 | A1–C1 | Machine | Read in Daily Life |
| Read an Academic Passage | 5–15 | B1–C2 | Machine | Academic Passage |
| Listening · 47 items · 35 raw points · adaptive — router 18 min, module 7–11 min | ||||
| Listen and Choose a Response | 15–19 | A1–B2 | Machine | Listen & Choose a Response |
| Listen to a Conversation | 10 | A2–C1 | Machine | Conversations |
| Listen to an Announcement | 6–10 | A2–C1 | Machine | Announcements |
| Listen to an Academic Talk | 8–16 | A2–C2 | Machine | Academic Talks |
| Writing · 12 items · 20 raw points · linear — 23 min | ||||
| Build a Sentence | 10 | A1–C2 | Machine | Build a Sentence |
| Write an Email | 1 | B1–C2 | AI | Write an Email |
| Write for an Academic Discussion | 1 | B1–C2 | AI | Academic Discussion |
| Speaking · 11 items · 55 raw points · linear — 8 min | ||||
| Listen and Repeat | 7 | A1–C2 | AI | Listen and Repeat |
| Take an Interview | 4 | A1–C2 | AI | Take an Interview |
Where the points actually are
Speaking earns 55 of the exam's 145 raw points from just 11 items — 5 raw points per item. Reading earns 35 points from 50 items — 0.7 points per item. Per item, Speaking is worth roughly 7× as many raw points as Reading, which is the single most quotable fact in the whole blueprint.
Machine vs. AI scoring
Only the Writing section's 2 constructed responses and the Speaking section's 11 items are AI-scored. Every other item on the exam — every Reading and Listening question, plus Writing's Build a Sentence items — is machine-scored: checked deterministically against a stored answer key, with no AI involved and no per-item cost. We describe the split section by section rather than as exam-wide totals, because the blueprint's own summary row doesn't reconcile with its section rows on this point.
| Section | AI-scored items | Machine-scored items |
|---|---|---|
| Reading | 0 | 50 |
| Listening | 0 | 47 |
| Writing | 2 | 10 |
| Speaking | 11 | 0 |
Stimulus specifications
What the text, audio, and prompts actually look like, per ETS’s spec — restated in our own words, not quoted from the source document.
Reading
- Easier texts run about 15–50 words; the most complex extend to around 200.
- Complexity is calibrated with ETS's TextEvaluator: roughly 800–1100 for medium texts (about US grade 10–11) and 1100–1360+ for harder ones (grade 11–12+).
- On a Flesch-Kincaid scale that corresponds to roughly 7.4–10.5 for medium and 10.5–20.0 for high difficulty.
- Texts stay on screen for the whole task, so you can re-read while answering.
- Formats range from signs, menus, emails and social posts to full academic passages.
Listening
- The shortest prompts are single utterances of no more than six stressed syllables.
- Mid-length audio typically runs 35–100 words and may involve more than one speaker.
- The longest monologic passages reach about 250 words.
- Audio is a mix of AI-generated voices and human voice actors, spanning US/Canadian, Australian and British varieties.
- Every recording plays once, and each is paired with a visual of the speaker or speakers.
Speaking
- Each task opens with a scenario introduction delivered both as audio and as text on screen.
- Visual context — schematic images, maps, forms, or photos of the speakers — accompanies the prompt.
- Repetition items grow from single clauses to longer sentences with compound verbs and relative clauses.
- Open-response items move from short factual answers toward opinion, explanation, prediction and narration.
What ETS calls each task
6 of the exam's 12 tasks carry a fuller, more literal name in ETS's own document than the short label we use elsewhere on this site. If you're matching the official term, here's the mapping.
Frequently asked questions
How many items are on the TOEFL 2026 test?
ETS's blueprint specifies 120 items across four sections, worth 145 raw points in total, over 1 hr 23 min – 1 hr 29 min of test time.
How are the raw points split across sections?
Reading 35, Listening 35, Writing 20, Speaking 55, for 145 raw points in total.
Which TOEFL 2026 tasks are graded by AI rather than a machine key?
Only the Writing section's 2 constructed responses and the Speaking section's 11 items are AI-scored. Every other item on the exam — every Reading and Listening question, plus Writing's Build a Sentence items — is machine-scored: checked deterministically against a stored answer key, with no AI involved and no per-item cost. We describe the split section by section rather than as exam-wide totals, because the blueprint's own summary row doesn't reconcile with its section rows on this point.
How many questions follow an Academic Talk?
ETS states a range of 8–16 items for the Academic Talks task type as a whole. The blueprint doesn't publish a fixed question count for an individual talk, so that section range is the figure we cite.
Does ETS use the same task names we do?
Mostly, but 6 of the 12 tasks carry a fuller official name in ETS's document than the short label used on this site — see the naming table above for the exact mapping.