OpenAI API pricing is pay-as-you-go: you are billed per million tokens of text sent in and generated out, and on OpenAI's pricing page (fetched 6 November 2026) the five models compared here run from $0.10 per million input tokens on gpt-6-luna to $10 on gpt-6-astra. It is the metered route into OpenAI's models, which matters for the AI tools covered in our guide to AI search optimisation in South Africa, and the token bill for a small workload can turn out smaller than the work of connecting it.

The pricing page is in dollars, lists dozens of models and says nothing about VAT or South African data rules. This post converts two realistic workloads into monthly rand bills with every input shown, then covers the costs that sit outside the price list. Rand figures use R16.42/USD, the exchange-rate assumption of the SA Digital Cost Index (SADCI) for August 2026. It is an index assumption, not a live rate.

Both workloads are working assumptions, not measurements of any real business. Replace the token counts with your own and the arithmetic still holds.

Quick Answer

OpenAI API pricing is per token, not per seat. On the 6 November 2026 price list, 10,000 customer replies a month cost about R49 on gpt-6-luna and about R985 on gpt-6.1-sol at R16.42/USD, before VAT questions and any batch discount. Across the five models in the table, model choice alone moves the bill by up to a factor of 100. Confirm current rates and your VAT treatment before you budget.

Planning an AI tool and unsure what it will really cost?

Send us the task you want automated and your monthly volume, and we will help you work out which of the cost levers below matter for it.

Get a free cost check

What Does the OpenAI API Cost in South Africa?

The OpenAI API cost for any South African business is the token price of the model you pick, converted at whatever rate your card or bank applies, because the page shows one USD price list and mentions no country-specific pricing.

OpenAI prices each model per million input tokens (what you send) and per million output tokens (what the model writes back). The table shows five models from the pricing page as fetched on 6 November 2026, with the monthly bill for the customer-reply workload described in the next section.

ModelInput per 1M tokensOutput per 1M tokensMonthly bill, customer-reply workloadBest For
gpt-6-luna$0.10$0.50$3.00 (about R49)High volume where unit cost decides
gpt-5.4-mini$0.75$4.50$24.75 (about R406)A middle step on an older model line
gpt-6.1-terra$1$5$30 (about R493)Mid-priced default to test first
gpt-6.1-sol$2$10$60 (about R985)Tasks where lower-priced models fail your tests
gpt-6-astra$10$50$300 (R4,926)Rare, high-value tasks only

Rand figures are the dollar bill multiplied by R16.42. The page lists many more models, plus embeddings at $0.02 per million tokens on text-embedding-3-small, and it does not rate which model suits which job. The "Best For" column is a budgeting suggestion, not a quality ranking: test each model on your own prompts before you commit.

The useful number in the table is the ratio. gpt-6-astra costs 100 times gpt-6-luna on both input and output, so in this example, model choice is the largest lever in openai api pricing. Volume and output length move the bill too.

How Do Tokens Turn Into a Monthly Rand Bill?

A monthly bill is tokens in plus tokens out, each multiplied by the model's per-million rate, so you can price a job before you build it. A common rule of thumb for English text is that 100 tokens is roughly 75 words, which means openai api tokens are easier to estimate than they sound.

Worked example: a customer-reply assistant

Assumptions (ours, not measured): 10,000 requests a month. Each request sends 1,500 tokens (instructions, a few knowledge-base passages and the customer's message) and receives 300 tokens back.

Volume: 10,000 × 1,500 = 15 million input tokens; 10,000 × 300 = 3 million output tokens.

On gpt-6-luna: 15 × $0.10 + 3 × $0.50 = $1.50 + $1.50 = $3.00, about R49 at R16.42/USD.

On gpt-6.1-sol: 15 × $2 + 3 × $10 = $30 + $30 = $60, about R985.

On the four models priced at 5x output to input, input and output cost the same amount in this example, because output tokens cost 5x as much while there are 5x fewer of them (1,500 against 300). On gpt-5.4-mini the output side is larger ($13.50 against $11.25). If your assistant sends long documents in and short answers out, input dominates; if it writes long replies, output does.

Takeaway: Price the Job Before the Model

A support assistant answering 10,000 customer messages a month costs about R49 on the cheapest model in the table and about R4,926 on the most expensive of the five in the table, a 100-fold gap at R16.42/USD. The volume assumption is the same in both cases, so the model choice, not the message count, sets the bill.

What Does Drafting Content With the API Cost?

Drafting content costs little in tokens per piece, because a 2,000-word draft is only a few thousand output tokens. At the same rule of thumb, 2,000 words divided by 0.75 words per token is about 2,700 tokens per draft.

Take 200 drafts a month, each with a 1,000-token prompt: that is 0.2 million input tokens and 0.54 million output tokens. On gpt-6.1-sol that comes to 0.2 × $2 + 0.54 × $10 = $5.80, about R95. On gpt-6-astra it is 0.2 × $10 + 0.54 × $50 = $29, about R476.

Token cost is not the real cost of content. Editing, fact-checking and brand voice take human time, and a model that drafts cheaply can still publish a wrong claim. Content intended to be cited by AI answers has to be accurate and original first, which is the subject of our guide on making content easier for AI to quote.

Using AI drafts but unsure they will be cited anywhere?

Send us two of your published pages and we will review how they are structured for answer engines to quote.

Request a page review

Which Costs Sit Outside the Price List?

Costs outside the standard token price are batch discounts, fast mode, regional processing uplift and tax, and each changes the bill. Each is stated on OpenAI's pricing page or the SARS VAT page, so check both before you sign off a budget.

Batch processing. OpenAI says batch pricing gives a 50% reduction on standard rates across most models. For work that can wait, such as the 200 monthly drafts above, gpt-6.1-sol falls from $5.80 to $2.90, about R48. Live customer replies need an immediate answer, so they would pay standard rates.

Fast mode. OpenAI charges 2x standard rates for fast mode (formerly priority processing). A live chat widget may justify it; a back-office job may not. Other metered APIs bill on the same pay-as-you-use logic, as our guide to Google Maps API pricing in South Africa shows.

Regional processing. OpenAI states that regional processing endpoints carry a 10% uplift for models released on or after 5 March 2026.

VAT. SARS states that VAT is levied at the standard rate of 15% on supplies by registered vendors and that it is "also payable on the importation of goods and on imported services."

SARS publishes separate guides on imported services and on electronic services supplied by foreign suppliers. Whether OpenAI charges you VAT, or you self-account for it, depends on how you are invoiced and whether you are a VAT vendor making taxable supplies. Ask your accountant before the first invoice, not at year-end.

What Happens to Customer Data Sent to the API?

Sending customer messages to a foreign API is a cross-border transfer of personal information, and section 72 of POPIA allows it only on specific grounds. The Act's conditions include that the recipient is subject to a law, binding corporate rules or binding agreement providing "an adequate level of protection", or that "the data subject consents to the transfer".

Further grounds cover transfers necessary for a contract, and transfers for the data subject's benefit where consent is impracticable but would likely be given. Whether a given protection counts as "adequate", and whether OpenAI is a third party or an operator in your set-up, are questions for a lawyer.

Section 72 applies to a responsible party sending information to a third party in a foreign country.

OpenAI's data-controls page states that "data sent to the OpenAI API is not used to train or improve OpenAI models (unless you explicitly opt in to share data with us)". It also states that abuse monitoring logs are "retained for up to 30 days, unless longer retention is required by law". Eligible customers can apply for zero data retention.

The same page lists regional processing in the United States, Europe (EEA and Switzerland) and the United Arab Emirates, and storage-only regions including Australia, Canada, Japan and India. South Africa is not listed.

For an openai api south africa deployment that handles customer personal information, that means your transfer basis under section 72 needs to be documented, ideally with a lawyer, before launch. Strip names, ID numbers and account details from prompts where the task does not need them.

Takeaway: Three Checks Before You Send Customer Data

Before customer messages go to OpenAI, confirm your VAT treatment with an accountant, document your POPIA section 72 transfer basis, and remove personal details the task does not need. OpenAI's data-controls page lists no South African processing or storage region.

Should You Pay per Token or per Seat?

A token-billed API suits automated, repeated or embedded tasks, while a flat-fee chat subscription suits a person doing varied work by hand. The API needs a developer or a no-code tool to connect it to your systems; a subscription needs only a login. Our guide to ChatGPT Plus in South Africa prices the subscription side in rand.

Your situationBetter fitWhyBest For
One person drafting and researching ad hocSubscriptionNo build work, no per-token trackingOwners and marketers
Thousands of repeated, similar tasksAPICost scales with volume and can use batch pricingSupport, tagging, product copy
AI inside your own site, app or workflowAPIThe API is the route for building AI into your own productDevelopers and agencies
Customer chat on WhatsAppAPI plus a messaging channelThe messaging channel is a separate bill from the modelSee our WhatsApp Business app versus API comparison

The token cost for a single occasional user is usually low, but the cost of building and maintaining the connection is not in the token price. Budget developer time separately.

Not sure whether AI will bring you customers or just cost you time?

Book an audit and we will show where your business already appears in AI answers and where it does not.

Book a free audit

Why South African Businesses Choose Growth Pulse Media

Growth Pulse Media is a Johannesburg agency founded by Dirk van Greuning, who built and scaled a large South African ecommerce business before starting the agency. We do not sell OpenAI integrations, and this post is written from OpenAI's and SARS's published pages, not from running a build on the API.

What we do cover is the visibility side: how AI answer engines describe and recommend your business. If that is your question, our AEO agency in South Africa page explains the work. All work is executed in-house and we keep a limited client load so each account gets senior attention. For related cost questions see AI lead generation in South Africa.

Who This Is NOT For

You need a fixed monthly figure. Token bills move with usage, model choice and the exchange rate. This post gives a method and two worked examples, not a quote.

You want a model recommendation. OpenAI's price list does not rate quality, and neither does this post. Test models on your own prompts.

You want legal sign-off. The POPIA and VAT sections point to the rules and sources. Your lawyer and accountant decide how they apply to you.

Frequently Asked Questions

How much does the OpenAI API cost in South Africa?

OpenAI prices per token in dollars, so the cost depends on the model and your volume. On the 6 November 2026 price list, 10,000 customer replies a month cost about R49 on gpt-6-luna and about R985 on gpt-6.1-sol at R16.42/USD, using our stated token assumptions.

Do I pay VAT on OpenAI usage in South Africa?

It depends on how OpenAI invoices you and whether you are a VAT vendor. SARS states that VAT at 15% is also payable on imported services, and it publishes separate guides for imported services and for foreign electronic-services suppliers. Confirm your treatment with an accountant before the first invoice.

Does OpenAI train on data sent through the API?

OpenAI's data-controls page says data sent to the API is not used to train or improve its models unless you explicitly opt in. Abuse monitoring logs are retained for up to 30 days unless longer retention is required by law, and eligible customers can apply for zero data retention.

Can I send customer personal information to OpenAI under POPIA?

Possibly, but sending personal information to a foreign provider is a cross-border transfer under section 72 of POPIA and needs a lawful basis, such as consent or a binding agreement providing adequate protection. South Africa is not among OpenAI's listed processing or storage regions, so document your basis and minimise the personal data in prompts.

Should I use the API or a ChatGPT subscription?

For one person doing varied work, a subscription is simpler and needs no build. For high-volume repeated tasks, per-token billing can total very little, as the R49 customer-reply example shows, but you must add developer time. Compare using your own volume.

Want AI to Send Customers to You, Not Just Answer Questions?

Tell us about your business and we will review how it shows up in AI answers across Google, ChatGPT and Perplexity. No obligation. We will get back to you within 24 hours.

Get Your Free Audit
Dirk van Greuning — Founder, Growth Pulse Media
Dirk van Greuning Founder, Growth Pulse Media

Founder of Growth Pulse Media and a specialist in South African search dominance. Dirk translates his experience in scaling South African businesses into high-velocity digital strategies for B2B and retail leaders. He writes about SEO, lead generation, and paid media from an operator's perspective — prioritising pipeline value over impressions.

Connect on LinkedIn