AI course creation cost is the total of what you pay model providers for tokens, images and audio, plus platform fees and the review hours needed to turn an AI draft into a publishable course. In the illustrative model below, the provider bill is under 1% of that total, yet it is the part you can cap, scope and log.
Authoring tools price AI either inside a seat or credit allowance, or through your own API key (BYO key) billed by the provider. This guide gives a cost formula, a worked example, spend controls and a governance checklist.
Disclosure: SimpliAuthor is a SimpliTrain product, made by the company that publishes this blog. It appears only near the end. Prices and policies were read from vendor pages in October 2026 and can change.
Key takeaways
- Tokens, retries, long sources, images, voice and translation drive usage spend, but review labor dwarfs it in our example.
- Credit plans give a predictable invoice. A BYO key gives control, but you must cap it yourself.
- Check the account tier before connecting a key, because Google treats unpaid and paid data differently.
What drives AI course creation cost?
AI course creation cost comes from seven drivers: tokens sent in and generated out, the model you choose, regeneration and retries, long source documents, images, text-to-speech and translation. Review labor sits on top, and it can dominate the total for an AI-assisted course.
All prices below are provider list prices as of October 2026 and can change.
- Tokens and model choice. Anthropic lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens, and Claude Haiku 5.5 at $0.10 and $0.50 for prompts up to 100,000 tokens. OpenAI lists gpt-6.1-sol and gpt-6-luna at the same two price pairs for standard short-context prompts. Output costs five times input.
- Retries and long sources. Every attempt is billed again, and a source document is input on every call that includes it. Anthropic says Claude 4.7 and later models produce about 30% more tokens for the same text, so token counts do not carry between models.
- Images and voice. Google lists gemini-nano-banana-2.1 at $0.0336 per 1K image, billed for every generated image, and gemini-3.8-flash-tts at the equivalent of $0.00225 per 10 seconds of audio.
- Translation. Each target language regenerates the text, as covered in course translation for international clients.
Some prices change on set dates: Google lists gemini-3.8-flash input at $0.75 per million tokens through December 31, 2026 and $1.50 from January 1, 2027.
How do bundled credits and bring-your-own-key pricing compare?
Bundled credits charge you through the authoring tool, as a seat price or subscription with an allowance. Bring-your-own-key tools bill your model provider for tokens. Credits give a predictable invoice. A BYO key gives control, but the invoice varies and you must cap it yourself.
The table shows four pricing models from vendor pages as of October 2026. We did not test these products.
| Pricing model | Example from the vendor page | What the page does not say |
|---|---|---|
| Seat, AI included | iSpring Suite AI: $1,290 per author per year | Any AI usage limit |
| Seat plus credits | Articulate 360 AI Teams: $1,749 per seat per year. Credits from $250 for 50,000 | Credits per course (“Credit usage varies by subscription and feature”) |
| Subscription with allowance | Mindsmith Professional: $39 per author per month, 2,000 AI credits | What a credit equals in tokens |
| Bring your own key | LearnDash Course Outline Builder needs your own OpenAI key. Its help page says 10 courses of 10 lessons cost $0.0038 on GPT-3.5-Turbo | A date for that figure (the page was updated April 22, 2026) |
How do you calculate AI course creation cost for one course?
AI course creation cost for one course equals text cost plus image cost plus audio cost plus review labor. Text cost for each step is runs times attempts times the token charge, where the token charge is input tokens times input price plus output tokens times output price, divided by one million.
Text cost per step = runs x attempts x ((input_tokens x input_price) + (output_tokens x output_price)) / 1,000,000 Image cost = images generated (kept + rejected) x price per image Audio cost = 10-second blocks rendered (including retakes) x price per block Total cost = text + images + audio + (review hours x loaded hourly rate)
Illustrative example, not a vendor price or a measurement. The course runs 60 minutes across 8 modules, with about 40,000 tokens of source. Token counts, attempts, review hours and the $50 hourly rate are our assumptions. Prices are the list prices above.
| Line item | Assumption | At $2 / $10 per million | At $0.10 / $0.50 per million |
|---|---|---|---|
| Outline | 3 attempts, 41,000 in, 3,000 out | $0.34 | $0.02 |
| Scripts | 8 modules x 2 attempts, 11,000 in, 2,500 out | $0.75 | $0.04 |
| Quizzes | 8 modules x 2 attempts, 3,000 in, 1,200 out | $0.29 | $0.01 |
| Translation | 1 language, 30,000 in, 36,000 out | $0.42 | $0.02 |
| Images | 30 generated at $0.0336 | $1.01 | $1.01 |
| Voice | 60 minutes plus 25% retakes | $1.01 | $1.01 |
| AI usage subtotal | 377,000 tokens in, 104,000 out | $3.82 | $2.11 |
| Review labor | 12 hours at $50 | $600.00 | $600.00 |
| Total | AI usage is about 0.6% of the total | $603.82 | $602.11 |
Switching text models saves about $1.71, while images and voice cost $2.02. Adding 40,000 source tokens to each of the 16 script calls would add $1.28.
Review labor is the number to manage, and the course development cost calculator shows how to model it. The formula omits platform fees and taxes, and data residency adds cost: Anthropic lists US-only inference at 1.1 times standard pricing.
Pro tip: Log Attempts, Not Just Spend. Record how many attempts each stage needed. The vendor sets the per-token price, but your prompts and source preparation can reduce attempts, and attempts multiply every text line.
Which controls keep AI course creation cost from drifting?
Three workflow controls and three provider controls keep AI course creation cost predictable. In your workflow, route simple tasks to small models, send excerpts instead of whole documents, batch repeated work, and cap attempts. At the provider, set monthly spend limits, scope keys, and turn on alerts.
What can you change in your own workflow?
- Route and excerpt. Use a small model for formatting and draft quiz distractors, a larger one for scripts, and pass each module only its source section.
- Batch. Anthropic and Google each list a 50% batch discount. Batch work is asynchronous, so it suits bulk translation, not live editing.
- Cap attempts. Allow two regenerations, then hand off to an editor.
Which spend controls do the providers document?
OpenAI, Anthropic and Google each document monthly spend controls, but they differ on how the limit is enforced, what error appears and how keys are scoped. The table summarizes their documentation as of October 2026. Confirm limits in each console.
| Provider | Spend control | When the limit is hit | Key scoping |
|---|---|---|---|
| Anthropic | Organization limit, capped by tier ($500 Start, $1,000 Build, $200,000 Scale per month). Workspace limits and alerts | Your own limit returns HTTP 400. The tier cap returns 429 until next month | Keys can be scoped to one workspace. The Default Workspace cannot have limits |
| OpenAI | Alerts that do not cap spend, plus hard limits per organization and project | HTTP 429. Spend can slightly exceed the limit | Limits apply per organization or per project |
| Google Gemini API | Project cap (labeled experimental), plus tier caps of $250, $2,000 and $20,000 to $100,000 | Project caps can overrun for about 10 minutes. A tier cap pauses all projects on the account | All keys in a project share its cap |
The OpenAI spend limits guide warns that hard limits can interrupt production traffic, and the Anthropic workspaces documentation says workspace limits cannot exceed the organization’s.
How do you govern a bring-your-own-key setup?
Governing a bring-your-own-key setup means deciding who owns the provider account, which data may be sent, how long the provider keeps it, who reviews output before publishing, and what gets logged. The key is both a budget decision and a data decision.
What do the providers say about your data?
OpenAI, Anthropic and Google each state that data sent through paid or commercial API access is not used to train their models, but retention and review rules differ. As of October 2026, their documents say the following.
- OpenAI: Its data controls guide says API data is not used to train its models unless you opt in, and abuse monitoring logs are kept for up to 30 days by default.
- Anthropic: Its Commercial Terms say Anthropic “may not train models on Customer Content from Services.” Its privacy center says API inputs and outputs are deleted within 30 days, with exceptions such as up to 2 years when its safety systems flag a chat.
- Google: The Gemini API terms say paid services do not use prompts to improve products. Unpaid services do, and human reviewers may read that content.
A key changes only who is billed. Ask the authoring vendor in writing what it stores.
What belongs on a BYO key governance checklist?
A bring-your-own-key governance checklist lists nine decisions to make and record before any key goes into an authoring tool. Each row names the decision and the evidence to keep.
| Item | Decide | Evidence to keep |
|---|---|---|
| Key owner | One named owner and a backup hold the account | Owner record |
| Key scope | One workspace or project per team or client | Key list with scope |
| Account tier | Paid or commercial terms only | Tier and data-use setting |
| Data rules | What may never be sent | Written classification |
| Retention | Accept the default or request zero retention | Dated provider terms |
| Budget | Alerts at 50% and 80%, hard limit at your ceiling | Console settings |
| Review gate | SME and instructional designer approve before publishing | Signed review per course |
| Logging | Who ran what, attempts, tokens, cost per stage | Monthly log |
| Offboarding | Rotate keys when a person leaves | Rotation date |
When is bring-your-own-key the wrong choice?
Bring-your-own-key is the wrong choice when volume is low, nobody can own billing and keys, you need a fixed monthly invoice, or a hard cap that halts work would hurt delivery. A credit plan is simpler in those cases, even if the unit price is higher.
Choose a BYO key when you produce many courses, want per-stage cost visibility, or have client contracts that dictate the provider. To pass AI costs to clients, see how to price eLearning development.
What does SimpliAuthor say about Bring Your Own AI?
SimpliAuthor’s published AI course authoring page says you can use platform AI or connect your own provider account. This section repeats only what that page states as of October 2026. SimpliAuthor is a SimpliTrain product, made by the company that publishes this blog.
The SimpliAuthor AI course authoring page states that you can “Connect OpenAI, Gemini, or other providers with your own account and API key.” It lists three benefits: “Align AI usage with compliance & governance,” “Manage cost on your own terms,” and “Track AI activity for visibility.”
The page does not state pricing, credit allowances, a full provider list, how provider data is handled, or what the tracking records. We did not test the feature, so ask before relying on those benefits.
What should you do next?
Run one pilot course through the formula before you choose a pricing model. Log attempts and tokens per stage, starting with the outline step in our guide to an AI course outline from trainer notes, then compare the result with a vendor’s credit price.
Set an alert and a hard limit on a test project, and fill in the checklist with the people who will own the key. If the pilot shows low volume or no key owner, choose a credit plan.
To see the Bring Your Own AI option described above, visit the SimpliAuthor AI course authoring page, or request a SimpliTrain demo and bring your budget questions. Both belong to the company that publishes this blog.
Frequently asked questions
How much does AI course creation cost per course?
It depends on the model, attempts and media. In our illustrative model, a 60-minute course used about $2 to $4 of AI usage and $600 of assumed review labor. As of October 2026, Mindsmith lists $39 per author per month and iSpring Suite AI lists $1,290 per author per year.
Is bring-your-own-key cheaper than AI credits?
Not necessarily. The credit pages we read do not define a credit in tokens, so public pages cannot settle it. A BYO key bills per token at list prices and may cost less at high volume. At low volume, key administration can cost more than it saves.
What is a token in AI course creation?
A token is the unit of text a model reads or writes, and providers bill input and output tokens separately. Token counts differ by model, so the same script can cost more on a newer model. Some authoring tools use their own units, such as one character per token for narration.
How do I stop an AI API bill from running away?
Set a hard monthly spend limit and an earlier alert at the provider, and use one workspace or project per team. OpenAI says spend can slightly exceed a hard limit, and Google says project caps can overrun for about 10 minutes, so set alerts well below your ceiling.
Do model providers train on content sent through an API key?
OpenAI says API data is not used for training unless you opt in. Anthropic’s Commercial Terms say it may not train models on customer content. Google says paid services do not use prompts to improve products. Google’s unpaid services do, and human reviewers may read that content. Check your tier first.



