In this article
Knowing that Claude Sonnet 5.5 and GPT-6.1 sol both charge $2 per million input tokens tells you nothing about what you'll pay at the end of the month. What matters is how many tokens your specific use case consumes, at what volume, and which model you choose for it. The same volume of tokens can cost more than ten times more depending on the model: in this article's chatbot, the same workload ranges from $33 to $660 a month.
This article takes current OpenAI, Anthropic and Google prices and does the maths for four real business scenarios: not per-token price tables, but what each one costs per month.
The four scenarios
For each one, we assume a realistic average case, not the best possible scenario:
- Customer support chatbot. 400 input tokens (question + system instructions) and 250 output tokens per conversation. Volume: 200,000 conversations/month.
- Coding agent. 3,000 input tokens (file context + instructions) and 800 output tokens per call. Volume: 10,000 calls/month, a small team using it throughout the day.
- Document analysis. 8,000 input tokens (a typical document) and 400 output tokens (the summary or extraction). Volume: 2,000 documents/month.
- Marketing content generation. 600 input tokens (brief) and 1,200 output tokens (the generated text). Volume: 1,000 pieces/month.
What each one costs per month
October 2026 prices, using a mid-tier model from each provider (Claude Sonnet 5.5, GPT-6.1 sol and Gemini 3.8 Flash), before caching or batch discounts.
| Scenario | Calls/month | Claude Sonnet 5.5 | GPT-6.1 sol | Gemini 3.8 Flash |
|---|---|---|---|---|
| Support chatbot | 200k | $660 | $660 | $248 |
| Coding agent | 10k | $140 | $140 | $53 |
| Document analysis | 2k | $40 | $40 | $15 |
| Content generation | 1k | $13 | $13 | $5 |
Gemini 3.8 Flash is at a promotional price until 31 December 2026; from 1 January 2027 the figures in this column double ($495, $105, $30 and $10).
What the table shows
Volume dominates, not complexity per call. The chatbot is the simplest of the four cases, with few tokens per interaction, yet it's by far the most expensive, simply because it has 20x more calls than the coding agent. Before optimising the model, it's worth understanding where the real volume is.
At this tier, Anthropic and OpenAI tie and Gemini comes in below in every scenario. Claude Sonnet 5.5 and GPT-6.1 sol have the same price ($2 / $10 per million tokens) and cost exactly the same in all four scenarios. Gemini 3.8 Flash comes in 62.5% below both (25% below from January 2027, when the promotion ends). This is consistent with what we showed in the per-token pricing comparison: the ranking between providers doesn't change by use case, only the scale of what's at stake.
Document analysis and content are cheap even at decent volumes. With short output and a moderate number of calls, the most expensive stays at $40/month (Claude Sonnet 5.5 and GPT-6.1 sol on document analysis). It's the high-volume chatbot and the coding agent with large context that deserve budget attention.
The effect of choosing the model tier
The table above uses the mid tier of all three providers. But for the support chatbot, the biggest bill of the four scenarios, the task is usually simple enough for an economy model. Redoing the maths with each provider's latest economy model:
| Provider | Economy model | Cost/month (chatbot) | Saving vs. mid tier |
|---|---|---|---|
| Gemini 3.5 Flash-Lite | $149 | 40% | |
| OpenAI | GPT-6 luna | $33 | 95% |
| Anthropic | Claude Haiku 5.5 | $33 | 95% |
Anthropic and OpenAI are the most extreme cases, and they tie here too. Swapping Claude Sonnet 5.5 for the new Claude Haiku 5.5 cuts the chatbot bill from $660 to $33 (each conversation here stays far below 100K input tokens, the limit for Haiku 5.5's lowest price), and swapping GPT-6.1 sol for GPT-6 luna makes exactly the same cut. Both are twenty times less, for a task that in most cases doesn't even need the larger model. It's the same point made in the article on how to calculate the cost of an AI project: the choice of model weighs more on the budget than the choice of provider.
Where caching comes in
The document analysis scenario benefits most from prompt caching, because the input (8,000 tokens) usually includes instructions and an output format that repeat across every document; only the document content itself changes. All three providers discount repeated input tokens: reading from cache costs 5% of the input price (a 95% discount) on Claude Sonnet 5.5 and GPT-6.1 sol, and 10% (a 90% discount) on Gemini 3.8 Flash, the three models in this table; Google adds an hourly storage cost. In a scenario where half the input repeats between calls, this can cut a significant share of the $15 to $40 a month for this use case, and the proportion only grows with volume.
How to apply this to your case
- Measure the real tokens of a typical call, not the best case. Use the provider's console or a token counter.
- Multiply by the expected monthly volume. This is where scenarios with few tokens per call can end up more expensive than "heavy" low-volume ones.
- Test the economy tier first, especially for high-volume tasks like the chatbot in this article; the difference can be an order of magnitude.
- Reserve the mid or top tier for tasks where response quality matters more than cost: complex code, business decisions, final content with no review.
To run these numbers for your own case, use the AI token cost calculator. The AI costs guide brings the rest of the topic together in one place.
Conclusion
The per-token price is the starting point, not the answer. What decides the bill at the end of the month is the combination of volume, tokens per call and the model tier chosen, and as this article shows, that combination can make the same provider's bill vary more than tenfold depending on the decision. Before choosing a provider, do the maths for your real use case, not for the price table.
The figures used are from October 2026, before caching and batch discounts. Always check the official prices before settling on an architecture.
From the token bill to the project cost
Tokens are the small part of an AI project's cost. Most of it sits in integrating with the systems you already have, preparing the data and changing how the team works. To do the full sum, see how much it costs to implement AI in a company, how to measure AI ROI and the automation ROI calculator.
Sources
Prices checked on 27 October 2026 on the providers' official pages:
Writes about AI consulting for operations and processes: how to map and simplify processes, apply AI where it has the most impact and measure the gains.