Tokens & Costs
Subscription vs API calculator: ChatGPT, Claude and Gemini
Find out whether a ChatGPT, Claude or Google AI plan is cheaper than paying for the API by the token, for the way you actually use it. Free, and it runs in your browser.
The model you’d use most in the app. $2 input · $10 output per 1M tokens.
≈ 80 tokens
≈ 400 tokens
Every message resends the conversation so far.
Per conversation, e.g. a pasted document.
Reasoning models think before answering; it’s billed as output.
- With prompt caching (best case)
- $3.07
- Per message, on average
- $0.00656
The API is cheaper. Your usage would cost about $3.94 a month, less than the cheapest plan, ChatGPT Go ($8). A plan still adds app features the API doesn’t, such as the chat app, voice and file uploads.
| Plan | Price | API costs the same at |
|---|---|---|
| ChatGPT Go | $8 | 41 messages/day |
| ChatGPT Plus | $20 | 102 messages/day |
| ChatGPT Pro 100 | $100 | 508 messages/day |
| ChatGPT Pro 200 | $200 | 1,016 messages/day |
| ChatGPT Pro 500 | $500 | 2,541 messages/day |
Steps
How to use the subscription vs API calculator
- Choose ChatGPT, Claude or Gemini, and the API model you’d use most.
- Start from an example, or enter how many messages you send a day and how many days a month.
- Adjust message and reply length, how long your conversations run, and any files you attach.
- For reasoning models, add the hidden reasoning tokens per reply.
- Read the verdict, then compare the break-even point for each plan.
Method
How it works
Chat subscriptions charge a flat monthly fee with usage limits. The API charges for exactly what you use, per million tokens. Which is cheaper depends almost entirely on how much you use it, so this calculator works out your monthly API bill from your habits and compares it with every plan.
How the API cost is worked out
An API request carries the whole conversation so far, because the model keeps no memory between requests. So for each conversation the calculator adds up every message: message one sends your text; message two sends message one, its reply and your new text; and so on. Files or documents you attach are sent with every message too. Each reply is billed as output, together with any hidden reasoning tokens, which reasoning models produce before answering. If a prompt grows past a model’s long-context threshold, the higher rate applies to that request.
The total is multiplied by your conversations per month, using the model’s current list price. With prompt caching, the part of each message that was already sent can be read from the provider’s cache at a much lower price, and only the new part is written (some providers charge extra for writing). The calculator shows this as a best case, because caches expire after a few idle minutes.
Reading the verdict
If the API is cheaper even without caching, it says so. If the cheapest plan is cheaper even with caching, a subscription wins. In between, it depends on whether caching works for you, which is more likely in an app you build than in occasional chats. The table shows the break-even point for every plan: the number of messages a day at which the API costs the same.
What the numbers don’t include
Plans have usage limits that providers describe in relative terms (“5x more usage”) rather than tokens, so the calculator can’t tell whether a plan’s limits would cover your usage. A plan also includes things the API doesn’t, such as the chat app, memory, voice and web search (and image generation on some plans), while the API lets you automate, integrate and pick any model. Plan prices are US prices checked on 2026-10-08; API prices are refreshed daily (last on 2026-10-09). To price an API workload in more detail, with batch discounts, use the LLM API cost calculator; to measure your real message length, use the AI token counter.
Examples
Worked examples
Monthly API cost for each usage example
| Usage | GPT-6.1 Sol | Claude Sonnet 5.5 | Gemini 3.1 Pro Preview |
|---|---|---|---|
| Casual chat10 short questions a day, every day | $1.12$1.00 cached | $1.12$1.00 cached | $1.28$1.13 cached |
| Daily work40 messages a working day, with pasted documents | $11.35$6.09 cached | $11.35$6.09 cached | $12.18$6.87 cached |
| Deep thinking15 hard questions a day to a reasoning model | $18.31$16.66 cached | $18.31$16.66 cached | $21.39$19.61 cached |
| Coding agentExample: 300 agent steps a day over a 30,000-token codebase context | $496.85$88.90 cached | $496.85$88.90 cached | $506.09$115.63 cached |
| Plans | ChatGPT Go $8, ChatGPT Plus $20, ChatGPT Pro 100 $100, ChatGPT Pro 200 $200, ChatGPT Pro 500 $500 | Claude Pro $20, Claude Max 5x $100, Claude Max 20x $200 | Google AI Plus $4.99, Google AI Pro $19.99, Google AI Ultra (5x) $99.99, Google AI Ultra (20x) $199.99 |
API list prices on 2026-10-09; plan prices (US, monthly billing) checked on 2026-10-08. “Cached” is the best case with prompt caching.
FAQ
Frequently asked questions
Is ChatGPT Plus cheaper than the API?
For light use, usually not. Ten short questions a day to GPT-6.1 Sol would cost about $1.12 a month on the API, against $20 for ChatGPT Plus. The plan wins for heavy daily use, long conversations, and features the API doesn’t include, such as the app, voice and image tools. Enter your own usage above to see where your break-even is.
Is Claude Pro worth it compared with the API?
It depends on how much you use it. Our “daily work” example (40 messages a working day with pasted documents) costs about $11.35 a month with Claude Sonnet 5.5 on the API, or $6.09 with prompt caching, compared with $20 for Claude Pro. Coding agents are a different story: our coding example would cost about $496.85 a month on the API without caching, which is why heavy Claude Code users often choose a Max plan.
Why do long conversations cost so much more on the API?
The API has no memory between requests, so every new message resends the whole conversation so far. The tenth message in a chat pays for the nine before it again, so cost grows much faster than the number of messages. Starting new conversations more often, or using prompt caching, keeps it down. Chat apps do the same work behind the scenes; the subscription just hides it.
What is prompt caching, and why is it called a best case?
When a request starts with the same text as a recent one, providers can reuse it from a cache and charge much less for that part. In a conversation, everything except your latest message and the last reply can come from the cache. The cache expires after a few minutes without use, so if you pause between messages you pay full price again. That’s why the calculator shows the cached cost as a best case.
Does a ChatGPT or Claude subscription include API access?
No. Chat subscriptions and the API are billed separately: the API is pay-as-you-go per token, with its own account and keys. Some plans include the provider’s own coding tools, for example Claude Pro and Max include Claude Code, but that is not general API access.
Are these prices right for my country?
Plan prices are US list prices, last checked on 2026-10-08. In many countries plans are sold in local currency at different prices, and taxes may be added. API prices are in US dollars per million tokens and are usually the same everywhere, plus any tax. Follow the plan links in the table to see the price where you are.
How accurate is the estimate?
It is as accurate as your inputs. Words are converted to tokens at OpenAI’s rule of thumb for English (about four tokens for every three words); code and other languages use more. For a precise figure, measure a typical message in the token counter, or check the usage page of your API account after a few days.
Related
Related tools
- AI Token CounterCount tokens for GPT, Claude, Gemini, DeepSeek, Qwen and more.
- LLM API Cost CalculatorEstimate per-request, daily and monthly API costs.
- AI Model ComparisonCompare prices, context windows and features across models.
- AI Model Pricing PagesSpecs, real costs and cheaper alternatives for popular models.
- Context Window CheckerSee whether your text fits each model's context window.
- GPU / VRAM CalculatorCheck how much VRAM a local model needs and which GPUs fit.