DeepSeek vs Claude vs GPT: Price per Message in ₹ (2026)
One typical AI chat message costs between ₹0.03 and ₹1.42 in India, depending on the model. GPT-5 Nano costs about ₹0.03 a message and Claude Haiku 5.5 about ₹0.04. GPT-5 Mini is about ₹0.13, Claude Sonnet 5.5 about ₹0.71 and Claude Opus 5.5 about ₹1.42. That means ₹10 buys anything from 7 to almost 400 messages.
AI companies publish prices per million tokens in US dollars, which tells most people nothing. This guide converts them into the three numbers that matter in India: rupees per message, messages per ₹10, and rupees per coding agent step.
How we count a "message"
A message here means 1,000 input tokens (your question plus some context) and 500 output tokens (the answer). That is roughly a long question and a three-paragraph reply. A token is about three quarters of an English word.
Rupee prices use the provider's published US dollar list price and Velona's published formula: dollar price × live USD/INR rate × 1.005 forex buffer × 1.07 platform fee. At an exchange rate of ₹94.61 on 7 November 2026, every dollar of API usage comes to about ₹101.74.
Price per message for every major model
| Model | USD per 1M tokens (input / output) | ₹ per message | Messages per ₹10 | Messages per ₹100 |
|---|---|---|---|---|
| GPT-5 Nano | $0.05 / $0.40 | ₹0.03 | 393 | 3,931 |
| Claude Haiku 5.5 | $0.10 / $0.50 | ₹0.04 | 281 | 2,808 |
| GPT-5 Mini | $0.25 / $2 | ₹0.13 | 79 | 786 |
| GPT-5.1 | $1.25 / $10 | ₹0.64 | 15 | 157 |
| Claude Sonnet 5.5 | $2 / $10 | ₹0.71 | 14 | 140 |
| Claude Opus 5.5 | $4 / $20 | ₹1.42 | 7 | 70 |
The gap between the cheapest and the most expensive model here is about 56 times. For everyday questions, the cheaper models are usually good enough, so the price difference is the main thing you notice.
What about DeepSeek?
DeepSeek V4 Flash is one of the cheapest strong models, but its price moves more than the others. DeepSeek charges more during peak hours, which fall in Indian working hours, and its rates changed several times in 2026. On Velona, V4 Flash was ₹9.31 per 1M input tokens and ₹18.61 per 1M output tokens on 16 October 2026, which is about ₹0.02 a message, or more than 500 messages for ₹10. At DeepSeek's own published peak rate in September 2026 ($0.30 / $1.20), the same message costs about ₹0.09. Check the live price on the DeepSeek V4 Flash page before you compare.
Output costs far more than input
Every model here charges more for the words it writes than for the words you send. On GPT-5 Mini and GPT-5.1, output costs 8 times input. On the Claude models it costs 5 times. So a long answer drives your bill much more than a long question. Setting a maximum answer length (max_tokens) is the simplest way to keep costs down.
Price per coding agent step
Coding agents such as Cline, OpenCode and Aider send your code again with every step, so they are input heavy. A typical step is about 30,000 input tokens and 1,000 output tokens.
| Model | ₹ per agent step | Steps per ₹100 |
|---|---|---|
| GPT-5 Nano | ₹0.19 | about 517 |
| Claude Haiku 5.5 | ₹0.36 | about 280 |
| GPT-5 Mini | ₹0.97 | about 103 |
| GPT-5.1 | ₹4.83 | about 20 |
| Claude Sonnet 5.5 | ₹7.12 | about 14 |
| Claude Opus 5.5 | ₹14.24 | about 7 |
A developer running 150 agent steps a day spends about ₹146 a day on GPT-5 Mini or ₹1,068 a day on Claude Sonnet 5.5. A common pattern is to run the agent on a cheap model and switch to a stronger one only for the step that keeps failing. Prompt caching can cut these costs further when the same code is resent.
How this compares to a ₹399 subscription
ChatGPT Go costs ₹399 a month in India once its free year ends. At the per-message prices above, ₹399 would buy about 3,100 GPT-5 Mini messages or about 15,700 GPT-5 Nano messages on a pay-per-use API. That is 100 GPT-5 Mini messages every day for a month.
A subscription still makes sense if you use the app's extra features every day, such as voice mode, image generation and file uploads. Pay per use makes sense if you ask a few questions a week, or if you are a developer who needs an API key anyway.
Which model should you start with?
- Quick questions, classification, translation: GPT-5 Nano or Claude Haiku 5.5, at a few paise a message.
- Everyday chat, study help and most apps: GPT-5 Mini, at about ₹0.13 a message.
- Coding agents on a budget: DeepSeek V4 Flash or Claude Haiku 5.5.
- Hard coding and reasoning: Claude Sonnet 5.5 or GPT-5.1.
- The strongest answer, cost second: Claude Opus 5.5.
Try every model with ₹10
On Velona, one account and one API key work with all of these models and 300+ others. You top up from ₹10 with any UPI app, every request shows its exact cost in rupees, and switching models is a one-word change in your code or chat.
Compare the models yourself with ₹10 by UPI. Create a free Velona account
Frequently asked questions
How much does one AI message cost in rupees?
For a message of 1,000 input and 500 output tokens, about ₹0.03 on GPT-5 Nano, ₹0.04 on Claude Haiku 5.5, ₹0.13 on GPT-5 Mini, ₹0.71 on Claude Sonnet 5.5 and ₹1.42 on Claude Opus 5.5, as of November 2026.
What is the cheapest AI model to use in India?
Among major models, GPT-5 Nano and DeepSeek V4 Flash cost about two to three paise per message. Claude Haiku 5.5 costs about four paise.
How many AI messages does ₹10 buy?
About 393 on GPT-5 Nano, 281 on Claude Haiku 5.5, 79 on GPT-5 Mini, 14 on Claude Sonnet 5.5 and 7 on Claude Opus 5.5.
Is DeepSeek cheaper than GPT?
DeepSeek V4 Flash costs about the same as GPT-5 Nano per message and much less than GPT-5 Mini. Its price varies with time of day and provider, so check the live rate.
How we calculated this
Dollar prices are the providers' published API list prices: OpenAI and Anthropic list prices as of October 2026, and DeepSeek's published rates from September 2026. Rupee figures apply Velona's published pricing formula at a USD/INR rate of ₹94.61 on 7 November 2026. Message and agent-step sizes are typical examples, not averages. No caching or batch discounts are included. Prices change, so check each model page for the live rate.
Disclosure: Velona publishes this guide and sells access to these models. If your card already works with a provider, paying them directly in dollars avoids Velona's 7% platform fee.
Related: DeepSeek price in India, Claude Sonnet 5 pricing in ₹ and OpenAI API price in India.