How to Use Claude Code in India Without a Claude Subscription: Gateway Setup and Cost per Step in ₹ (2026)
Editor note before publishing: the commands in the setup section follow the Claude Code and LiteLLM documentation, but this exact chain has not been run end to end against Velona. Test it once and remove this note.
Yes, Claude Code can run without a Pro or Max plan. Claude Code's documentation describes a gateway mode: you point the ANTHROPIC_BASE_URL setting at a gateway that speaks the Anthropic Messages format, and Claude Code sends its requests there.[1] With a small local translator in between, those requests can be billed per token from a rupee wallet. One agent step then costs from ₹0.35 to ₹14.20, depending on the model.[6]
Key findings
- Claude Code is included in Claude Pro, reported at ₹2,399 a month in India.[4][5]
- Gateway mode is an official, documented feature and needs only environment variables.[1][2]
- Claude Code uses the Anthropic Messages format, so the base URL is the gateway root and not a path ending in
/v1.[3] - A developer running 150 steps a day for 22 days spends about ₹1,171 on Claude Haiku 5.5 and about ₹23,422 on Claude Sonnet 5.5 at catalogue prices.[6]
- For heavy daily use of Sonnet or Opus, the subscription is the cheaper route.
Scope and method
Setup steps are taken from the Claude Code gateway documentation and two independent setup guides.[1][2][3] Costs use Velona catalogue prices on 14 November 2026.[6] A coding agent step is counted as 30,000 input and 1,000 output tokens with no prompt caching, which makes the figures upper bounds. Subscription usage limits are not published as fixed step counts, so the comparison with Pro is a guide and not a measurement.
How the request travels
Claude Code on your laptop
| Anthropic Messages format
v
LiteLLM proxy on localhost:4000 (translates the format)
| OpenAI compatible format
v
https://velona.in/v1 (rupee wallet, UPI top-up)
|
v
The model you chose: Claude, DeepSeek, Qwen, Kimi or GPT
Claude Code's documentation uses LiteLLM as its worked example of a gateway and shows the Anthropic format endpoint.[1] Velona's public endpoint is OpenAI compatible, which is why the translator sits in the middle.
Cost per step in rupees
| Model | One agent step | Steps per ₹100 | 150 steps a day, 22 days |
|---|---|---|---|
| Claude Haiku 5.5 | ₹0.35 | about 281 | ₹1,171 |
| Qwen3 Coder Next | ₹0.45 | about 224 | ₹1,472 |
| DeepSeek V4.1 Flash | ₹1.03 | about 96 | ₹3,413 |
| Claude Sonnet 5.5 | ₹7.10 | about 14 | ₹23,422 |
| Kimi K3.1 Pro | ₹7.10 | about 14 | ₹23,422 |
| Claude Opus 5.5 | ₹14.20 | about 7 | ₹46,844 |
Source: Velona catalogue, 14 November 2026.[6] Claude Haiku 5.5 is priced for prompts up to 100,000 tokens, and longer prompts cost more.[7]
When the subscription is cheaper
The table makes the trade-off plain. A full working month on Sonnet 5.5 at pay per token prices costs about ten times the reported ₹2,399 Pro price. Plans have usage limits, so the two are not identical, but a daily Sonnet or Opus user should start with the subscription. Pay per token makes sense in four cases:
- You use Claude Code a few hours a week, not all day.
- You are happy to run most steps on a budget model and switch up only for hard problems.
- You cannot pay for the subscription because the website does not take UPI. See every way to pay for AI with UPI.
- You want one bill in rupees across several models.
Setup in five steps
1. Get a Velona key
Create an account at velona.in, top up with UPI and create an API key with a budget.
2. Install LiteLLM
pip install "litellm[proxy]"
3. Write the proxy config
# config.yaml
model_list:
- model_name: claude-sonnet-5.5
litellm_params:
model: openai/anthropic/claude-sonnet-5.5
api_base: https://velona.in/v1
api_key: os.environ/VELONA_API_KEY
- model_name: deepseek-v4.1-flash
litellm_params:
model: openai/deepseek/deepseek-v4.1-flash
api_base: https://velona.in/v1
api_key: os.environ/VELONA_API_KEY
4. Start the proxy
export VELONA_API_KEY=your_velona_key
litellm --config config.yaml --port 4000
5. Point Claude Code at it
export ANTHROPIC_BASE_URL=http://localhost:4000
export ANTHROPIC_AUTH_TOKEN=any-local-value
export ANTHROPIC_MODEL=deepseek-v4.1-flash
claude
The base URL is the proxy root with no /v1 at the end, because Claude Code adds its own path.[3] To keep the settings between sessions, the same values can go in the env block of Claude Code's settings.json.[2] If you were signed in with an Anthropic account earlier, one gateway guide advises running /logout once so cached credentials do not conflict.[8]
Things that behave differently on a gateway
- Tool calling quality varies by model. Claude Code was built around Claude models. Other models can run it, but expect more failed edits on weaker ones.
- Caching may not carry over. The cost table assumes no caching for that reason.
- Your key is your spending limit. An agent can loop. Set a budget on the key before a long run.
Limitations
- The setup follows published documentation and has not been verified end to end on Velona by us.
- Step size is an example. Real steps range from a few thousand to several hundred thousand tokens.
- We do not compare output quality between models in this guide. See how to evaluate coding models on your own code.
Frequently asked questions
How much does Claude Code cost in India?
It is included in Claude Pro, reported at ₹2,399 a month on the web.[4][5] On pay per token billing, one typical step costs ₹7.10 on Claude Sonnet 5.5 and ₹0.35 on Claude Haiku 5.5 at Velona prices.[6]
Can Claude Code use models that are not Claude?
Yes, through a gateway that accepts the Anthropic Messages format and forwards to another model.[1]
Is this allowed?
Gateway configuration is documented by Anthropic for Claude Code.[1] Each model provider's own terms still apply to how you use its model.
Do I need an international card?
Not for the pay per token route described here. The wallet is topped up with UPI.
Disclosure: Velona publishes this guide and is the gateway used in the example. Prices were checked on 14 November 2026.
References
- Claude Code documentation: LLM gateway configuration
- TrueFoundry: Claude Code with LiteLLM setup guide
- DEV Community: configuring Claude Code with a third-party API
- Croma Unboxed: Claude AI India pricing plans confirmed
- ExplainX: Anthropic Claude INR pricing in India, July 2026
- Velona: live model pricing in INR
- Velona: Claude pricing for every model
- Opper documentation: Claude Code gateway guide