
Claude Code Cost in Rupees: Estimate and Cap It
Claude Code sends many requests per task and resends a growing context each time, so input tokens drive the bill. Here is how to estimate Claude Code cost in rupees before you start, and how to cap it with a per-key budget.
Author
AICredits Team
Published
16 Nov 2026
Reading time
7 min read
Why Claude Code cost in rupees is hard to guess
Claude Code is a command-line coding agent. You give it a task, and it reads files, runs commands, edits code and checks the result. Each of those steps is a separate request to the model. A chat message costs one request. A coding task can cost dozens.
That is why the first question from most developers in India is the same: what will a session cost, in rupees, before I start? Paying in dollars adds a second unknown, because the bank rate and any forex charge decide the card statement. A prepaid rupee wallet removes that part. The token arithmetic is still yours to estimate.
You can run AICredits as Claude Code API credits India developers pay for in rupees. Top-ups start at ₹50, and you can pay by UPI, net banking or card. The rest of this post covers the cost model, three worked scenarios, how to pick a model, and how to set a hard cap.
Why agent sessions cost what they cost
A language model does not remember earlier turns. Every request carries the full conversation so far: your instructions, the model's earlier replies, and the output of every tool it ran. When Claude Code reads a 500-line file, that file becomes part of the context. On the next turn it is sent again, and again after that.
Three consequences follow:
- Input tokens accumulate. Turn 1 might send a few thousand tokens. Turn 30 sends everything from turns 1 to 29 as well, plus new tool output.
- Output is small by comparison. The model mostly writes short tool calls and small edits. Per turn, output is usually hundreds of tokens, while input is tens of thousands.
- Input price matters more than output price. Output tokens cost more per token, but there are far fewer of them. In the scenarios below, input is over nine-tenths of the bill.
Prompt caching softens this where the model supports it. Context that repeats from one request to the next can be billed at a lower cached-input rate. At November 2026 prices, cached input costs ₹19.82 per 1M tokens for Claude Sonnet 5.5 against ₹198.24 for uncached input, and ₹0.99 for Claude Haiku 5.5 against ₹9.91. How much of a real session gets cached depends on the session, so treat caching as a possible saving and not a promise.
The formula for estimating Claude Code cost in rupees
For one session:
cost = (turns × avg input tokens per turn × input price ÷ 1,000,000)
+ (turns × avg output tokens per turn × output price ÷ 1,000,000)Prices are in rupees per 1M tokens. You need three guesses: how many turns the task takes, how large the context is on average across those turns, and how much the model writes per turn. The cost calculator does the same arithmetic if you would rather not use a spreadsheet.
Three illustrative scenarios
The token counts below are assumptions for illustration. They are not measurements, and they are not a claim about what a typical session costs. Your sessions will differ with codebase size and how much the agent reads. Run one real session, then check the token counts and rupee cost in your dashboard and replace these assumptions with your own.
Prices used, at November 2026 prices, per 1M tokens: Claude Sonnet 5.5 is ₹198.24 input and ₹991.20 output. Claude Haiku 5.5 is ₹9.91 input and ₹49.56 output.
| Scenario | Turns | Avg input per turn | Avg output per turn | Total input | Total output |
|---|---|---|---|---|---|
| A. Small bug fix in one file | 12 | 20,000 | 400 | 240,000 | 4,800 |
| B. Feature across several files | 40 | 40,000 | 600 | 1,600,000 | 24,000 |
| C. Long refactor | 100 | 70,000 | 800 | 7,000,000 | 80,000 |
The arithmetic, with no caching:
| Scenario | Claude Sonnet 5.5 | Claude Haiku 5.5 |
|---|---|---|
| A | 0.24 × 198.24 = ₹47.58 input, 0.0048 × 991.20 = ₹4.76 output, total about ₹52.34 | 0.24 × 9.91 = ₹2.38 input, 0.0048 × 49.56 = ₹0.24 output, total about ₹2.62 |
| B | 1.6 × 198.24 = ₹317.18 input, 0.024 × 991.20 = ₹23.79 output, total about ₹340.97 | 1.6 × 9.91 = ₹15.86 input, 0.024 × 49.56 = ₹1.19 output, total about ₹17.05 |
| C | 7 × 198.24 = ₹1,387.68 input, 0.08 × 991.20 = ₹79.30 output, total about ₹1,466.98 | 7 × 9.91 = ₹69.37 input, 0.08 × 49.56 = ₹3.96 output, total about ₹73.33 |
Now add caching to scenario B on Claude Sonnet 5.5. Assume 80% of the 1,600,000 input tokens are billed at the cached rate. That is 320,000 uncached tokens at ₹198.24 per 1M, which is ₹63.44, plus 1,280,000 cached tokens at ₹19.82 per 1M, which is ₹25.37, plus the same ₹23.79 of output. The total is about ₹112.60 instead of ₹340.97. The 80% is another assumption; your share could be higher or lower.
The shape is what matters: turns multiply context size, so a longer session costs disproportionately more, and a smaller model changes the bill by an order of magnitude.
Live Claude prices
The table below is read from the live catalogue, so it stays current when new Claude models ship. Prices are in rupees per 1M tokens.
| Model | Input / 1M tokens | Output / 1M tokens |
|---|---|---|
Claude Opus 5.5anthropic/claude-opus-5.5 | ₹390.13 | ₹1,950.67 |
Claude Sonnet 5.5anthropic/claude-sonnet-5.5 | ₹195.07 | ₹975.34 |
Claude Haiku 5.5anthropic/claude-haiku-5.5 | ₹9.75 | ₹48.77 |
Claude Fable 5.2anthropic/claude-fable-5.2 | ₹780.27 | ₹3,901.34 |
Live prices in ₹ from the AICredits catalog. They follow the USD/INR rate, so they can differ from day to day. See all Anthropic models or the cost calculator.
Choose the model for the job
- Claude Sonnet 5.5 is the default for everyday coding: bug fixes, features, tests, code review.
- Claude Haiku 5.5 suits light edits, renames, formatting, boilerplate and quick questions about a file. As scenario A shows, it costs a small fraction of Sonnet for the same token volume.
- Claude Opus 5.5 is for the hardest work: tangled bugs, large refactors, and tasks where a failed attempt costs you more than the price difference.
Switching is one line. To run a single session on Opus without changing your saved settings:
ANTHROPIC_MODEL=anthropic/claude-opus-5.5 claudeCap spend with a dedicated API key
A hard limit beats an estimate. Create a separate AICredits API key just for Claude Code and give it its own budget.
- Open the dashboard and create a new API key for Claude Code.
- Set a budget limit on that key, for example ₹500.
- Use that key only in Claude Code, not in your applications.
When the key reaches its budget, requests on it stop. A session that loops, or a long refactor that runs further than you planned, ends at the cap and does not keep spending. Your wallet balance is the overall ceiling above that: no key can spend more than you have topped up.
The usage dashboard shows cost per key and per model, so you can see what a session cost and compare it against the scenarios above.
Set up Claude Code with AICredits in three steps
Full instructions, including shell-export alternatives, are in the Claude Code setup guide. In short:
- Install Claude Code (
npm install -g @anthropic-ai/claude-code, Node.js 18 or newer) and top up your AICredits wallet. - Create an API key in the dashboard and set a budget on it. The key starts with
sk-and is shown once. - Add the settings to
~/.claude/settings.json:
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.aicredits.in",
"ANTHROPIC_AUTH_TOKEN": "sk-your-aicredits-key-here",
"ANTHROPIC_MODEL": "anthropic/claude-sonnet-5.5"
}
}Use https://api.aicredits.in without /v1 for Claude Code, and use ANTHROPIC_AUTH_TOKEN, not ANTHROPIC_API_KEY. Open a new terminal before running claude, because the variables are read once at startup. The guide lists what to check for each Claude Code feature, along with fixes for 401 and 402 errors and for duplicate requests. For more on the Claude side, see Claude API in India.
After your first prompt, check the usage page. The request and its rupee cost should show up within a few seconds.
Frequently Asked Questions
Can I pay for Claude Code with UPI?
Yes. You top up the AICredits wallet through Razorpay using UPI, net banking or a card, with a minimum of ₹50. Claude Code with UPI works because the payment goes into the wallet, and each request is then charged from the wallet balance in rupees.
Does this use my Claude subscription?
No. This is API usage, billed per token from your AICredits wallet. It is separate from any Claude subscription you may have, and it does not draw on one.
How do I stop a runaway session?
Give the Claude Code key a budget so requests stop at the cap. When the wallet or the key runs out, requests fail with a 402 error until you add credits or raise the limit.
Do I get a GST invoice?
Yes. A GST invoice is issued for every top-up. GST at 18% is included in the amount you pay. Add your business name and GSTIN to your billing profile before you top up so they appear on the invoice.
Get started
- Create an account and top up from ₹50.
- Follow the Claude Code setup guide to point Claude Code at your key.
- See live rupee prices and what the Claude models are good for on Claude API in India.
AICredits
Build with any AI model. Pay in INR.
One OpenAI-compatible API for every major model, with a prepaid rupee wallet. Top up via UPI — no international credit card, no forex surprises.
Top up via UPI, prepaid INR wallet, no international card needed
Related articles
Continue in docs
Need implementation commands and endpoint details? Go to quickstart or API reference.