ChatGPT API Pricing 2026: What the OpenAI API Actually Costs

Short answer first: there is no separate product called the “ChatGPT API,” and no separate price list for it. ChatGPT is the subscription chat product. The API that tutorials, plugins and startups mean when they say “ChatGPT API” is the OpenAI API — you get a key from the OpenAI platform, pick a model per request, and pay per token. That billing is completely separate from a ChatGPT subscription: a Plus plan buys you chat usage limits, not API credit, and an API balance buys you tokens, not a ChatGPT seat. (The closest thing to a literal “ChatGPT model” on the API is chat-latest — covered below.)

So the real question is what the OpenAI models people use for ChatGPT-style apps cost. The spread is wide — the same workload costs $1.80 or $180 a month depending on the model you pick — which is why this page exists. For the complete 14-model ladder with every cached and batch rate, see OpenAI API pricing; this page covers the shortlist most “ChatGPT API” searches are actually about, plus the worked costs.

Disclosure: this article contains no affiliate links. Where CostMyTokens does use affiliate links elsewhere, they never affect order — every table is sorted by price only. Read our disclosure.

Prices are OpenAI’s official API list prices from its official pricing page, re-checked 2026-10-28, and match the verified dataset behind our OpenAI provider page. Standard workload for monthly figures: 10M tokens, split 8M input / 2M output, no caching, standard service tier.

The shortlist: models people mean by “ChatGPT API”

How to read this as a buyer, not a catalog:

• Building a chatbot or assistant on a budget: the Luna tier is the default starting point — GPT-6 Luna at $0.1 / $0.5 is the cheapest way to serve chat traffic, and its cached input ($0.01) makes repeated system prompts nearly free.
• The mainstream default: GPT-6.1 Sol at $2 / $10 is the tier most general-purpose apps land on when Luna’s answers are not good enough. If Luna falls short but Sol feels like overkill, GPT-6.1 Terra sits in between at $1 / $5 (cached $0.05) — exactly half of Sol, $18.00 at the standard workload. Note that GPT-6 Sol carries the same sticker price but a cached-input rate twice as high ($0.2 vs $0.1) — on cache-heavy traffic the 6.1 is the cheaper Sol.
• The frontier tier: GPT-6 Astra at $10 / $50 costs 100x Luna per input token. It is priced for the hardest agentic and coding work, not for chat volume.
• The literal “ChatGPT model”: chat-latest points to the Instant model currently used in ChatGPT, at $5 / $30 — pricier than GPT-6.1 Sol on both columns, with no Batch support and a snapshot that changes under you. OpenAI’s own model page recommends GPT-6 Astra for production API use instead. Calling it gets you ChatGPT’s current behavior, not a discount.
• One date to watch: OpenAI says GPT-5.6 Sol’s promotional pricing ($4 / $20) is available at least through 2026-11-21. Budgets built on it should be re-checked after that date.

Six more models (GPT-4o, GPT-4o Mini, GPT-4.1 Mini, GPT-6.1 Terra, GPT-5.6 Terra, GPT-6 Sol) are in the full OpenAI ladder. None of them changes the shape above: the ladder still runs from $1.80 to $180 at the standard workload.

What an actual app costs

Per-token prices are hard to budget from, so here are two complete workloads, computed at list price with no caching:

A support chatbot — 10,000 conversations a month, 2,000 input and 500 output tokens each:

A coding agent — 1,000 sessions a month, 50,000 input and 5,000 output tokens each:

Caching moves these numbers materially when your prompt is stable. If 80% of the support bot’s input is a reused system prompt served from cache, the GPT-6.1 Sol bill falls from $90.00 to $59.60 a month, and Luna’s from $4.50 to $3.06 (cache-read savings only; OpenAI also lists cache-write rates on the newest models, which a stable prompt pays rarely). The prompt caching guide works through that math in full.

ChatGPT subscription vs the API, in one table

If you are deciding between the two for your own use rather than for an app, the token math for a $20 budget is in subscription vs API: the break-even, in tokens.

Three billing traps for first-time API buyers

• There is no free API tier. OpenAI does not publish a free token allowance for the API — usage is billed per token, which is why “ChatGPT API free” searches end in disappointment or in unofficial resellers. How OpenAI compares on free access is covered in free LLM API tiers compared.
• Speed is a multiplier, not a setting. On top of standard rates, OpenAI sells Fast and Ultrafast service tiers; Ultrafast Astra is $60 / $300 per 1M — six times standard. Check which tier your keys are on before scaling traffic.
• Long conversations reprice. Prompts above roughly 272K input tokens enter a long-context segment where input and cached rates roughly double. A support bot whose threads grow unbounded will drift into it quietly. Batch, where your work can wait, halves the input and output rates of every model in the table above except chat-latest, which does not support Batch — see batch API discounts explained.

FAQ

Is there a ChatGPT API key? Not as a separate product. You create an OpenAI API key on the OpenAI platform, and that key can call the models in the table above. Keys shown inside ChatGPT-related tutorials are OpenAI API keys.

Does ChatGPT Plus include API access? No. The subscription and the API are billed by different systems; Plus gives you chat usage, and API calls are charged to your API balance regardless of any subscription you hold.

Which model does ChatGPT itself use — and is that the one I should call? ChatGPT the product routes between models and changes over time. The API exposes its current Instant model as chat-latest ($5 / $30 per 1M), but OpenAI recommends a fixed model such as GPT-6 Astra for production, and a pinned model is easier to budget. On the API you name the model in every request, and your bill follows your choice — which is exactly why the $1.80-to-$180 spread exists.

Run your own numbers

The examples above use fixed conversation shapes. Your input/output split, cache hit rate and volume will move the answer — plug your real numbers into the CostMyTokens calculator, or open any model page in the table to cost that model on its own. The complete OpenAI catalog, ranked, is in OpenAI API pricing 2026. For how xAI’s price list compares — including a flagship that undercuts GPT-6.1 Sol on output — see Grok API pricing.

Prices: OpenAI’s official API pricing, re-checked 2026-10-28 and tracked on our OpenAI provider page with per-model verification dates shown in the table. Prices change; check the official page before committing budget.