t·Tuition HelpSign in

Typical costs: a 20-child lesson and how costs grow

Illustrative costs from the same calculator used by the app, with clear assumptions.

For: Teacher, Centre Manager, Super Admin

  1. These are planning examples, not a promised invoice or the current settings of your centre. They use reviewed standard prices, exclude taxes/hosting and do not make AI calls.
  2. Start with 20 learners for 20 minutes and one tutor reply per minute. The comparison below includes the configured example checker for text and recorded voice.
  3. The written maths row assumes 20 submitted answers per child, with 1,500 input and 250 output tokens per answer using GPT-4.1 mini. It is separate from chatting; actual marking cost varies with the question and response length. Teacher approval and overrides do not call AI.
  4. For live voice, the example uses 15 seconds of child speech and 20 seconds of tutor speech per reply. Tutor seconds include every repeated language. Both cache and no-cache scenarios are shown; caching is not guaranteed.
  5. More replies, longer explanations, more spoken languages, bigger models and more reasoning can increase cost. Repeated conversation history also contributes to live-call cost.
  6. A quiet connection alone has no connection fee, but audio included in responses and reused context can be billable. Pauses still use connected call time and the app’s time limits.
  7. Use saved course audio for repeated questions, shorter replies and prepared practice where useful. Preserve the voice support the learner needs; measure a small real lesson before changing class-wide limits.

20 minutes each, 20 children

All amounts below are US dollars. Budget settings use US cents: 100 cents = $1. These examples use 20 replies per child. Text uses GPT-4.1 mini with a GPT-5.4 mini checker; your centre may choose different models.

Planning estimates, including the conversation assumptions below
How the child learnsOne child20 children
Typed chat + checker$0.18$3.61
Written maths marking · 20 answers per child$0.02$0.40
Recorded Quick Voice + text + checker + spoken reply$0.28$5.65
GPT-Realtime-2.1 mini · lower cost live voice · no cache assumed$0.86$17.18
GPT-Realtime-2.1 mini · lower cost live voice · cache scenario$0.36$7.12
GPT-Realtime-2.1 · improved noise handling live voice · no cache assumed$2.83$56.61
GPT-Realtime-2.1 · improved noise handling live voice · cache scenario$1.19$23.85
GPT-Realtime-2 · Playground option live voice · no cache assumed$2.83$56.61
GPT-Realtime-2 · Playground option live voice · cache scenario$1.19$23.85

These are scenarios, not a reliable minimum or maximum. Text assumes 3,000 input and 300 output tokens per reply, plus checker input of 5,000 and output of 800 including reasoning. Quick Voice adds Whisper-1 transcription and TTS-1 speech at about 12 characters per second. Live voice assumes 1,500 instruction tokens, up to 4,000 history tokens, 15 seconds of child speech and 20 seconds of tutor speech per reply. It charges repeated history conservatively as audio. The cache scenario discounts 80% of repeated audio history; cache hits are not guaranteed. Live speech is immediate and does not include a separate pre-speech correctness checker.

Rates and billing behaviour: OpenAI API pricing and Realtime usage and costs. Reviewed 10 September 2026. Taxes, hosting and other class activity are excluded.

How a small cost grows

  • More children: ten identical lessons cost half as much as twenty. A centre allowance is shared across the whole class.
  • More back-and-forth: at two replies per minute, with 10 seconds of child speech and 15 seconds of tutor speech per reply, this calculator estimates $10.18 for a 20-child Quick Voice class. Typed chat with checks becomes $7.22.
  • More languages: if English takes 20 seconds and Tamil adds 20 seconds, enter 40 tutor-speaking seconds in the calculator. Extra speech and later repeated history both matter; doubling speech does not simply double the entire bill.
  • Longer answers and stronger models: a longer explanation, more reasoning or a more expensive checker can increase cost even when the number of replies stays the same.
  • Pictures and recordings: generating or checking a new item costs money. Replaying an existing saved item does not make another AI call.

Illustration allowance example

With the example checker, the app currently reserves about $0.10 for low quality, $0.15 for medium quality, $0.31 for high quality per new illustration, including its check. These are bounded reservations, not the final image prices. Complete usage releases any unused money; incomplete usage stays held. High quality and a checker still cannot guarantee a mathematically correct picture.

A practical centre budget example

  1. Open AI cost insights. Set 20 children and 20 minutes, select realistic speaking and reply amounts, and read both cache and no-cache estimates.
  2. For the mini live-voice example above, the no-cache class estimate is $17.18. Twenty simultaneous replies also need $1.60 of temporary free headroom. A $1 shared voice ceiling will stop this lesson well before 20 minutes each.
  3. For illustration only, a manager might test a $25 shared centre limit and $25 voice limit, with $1.25 shared and voice limits per child. Enter those as 2,500 and 125 cents. This gives more headroom for this scenario, but is not a guarantee or a recommended setting for every centre.
  4. Check the remaining limits too: overall and voice operation counts, live minutes/day, maximum minutes/call, student eligibility and learning time. Twenty planned replies need at least twenty available operations; leave room for the other work you intend to allow. A shorter call cap still splits the lesson into shorter calls.
  5. Run a short supervised lesson, then compare Estimated actual spending, Reservations awaiting settlement and Unused reservation released. Use observed results before opening the same allowance to the full class.

These examples do not change settings or purchase credits. Open Budgets at the top for a today-only top-up. Centre Managers can approve increases within the Super Admin ceiling. Teachers can request an increase, or apply it within an allowance explicitly enabled by a Super Admin. Only a Super Admin can raise the period ceiling.

This guide explains the feature; reading it does not change any setting or grant access. Back to all help