The Complete LLM API Price List in Rupees (2026)

Why this list exists

Every provider publishes prices in USD per million tokens. If you are budgeting in rupees, that means a mental forex calculation every time you compare two models — and the number moves with the exchange rate. This is the LLM API price list with everything already converted to ₹, using the same forex-plus-buffer conversion AICredits applies to live wallet deductions.

Prices below use an effective rate of ₹87/USD with a 5% forex buffer, matching AICredits' published rates on individual provider pages. Treat this as a reference point — check aicredits.in/calculator for the current live rate against your exact workload, since forex moves and provider prices change.

Chat models — cost per million tokens

| Model | Input (₹/1M) | Output (₹/1M) | Provider | |-------|--------------|----------------|----------| | GPT-4o-mini | ₹13.23 | ₹52.92 | OpenAI | | Gemini 1.5 Flash | ₹7.22 | ₹28.88 | Google | | Gemini 2.0 Flash | ₹9.63 | ₹38.51 | Google | | DeepSeek V3 | ₹12.18 | ₹48.72 | DeepSeek | | Mistral Small | ₹19.26 | ₹57.78 | Mistral | | Claude 3.5 Haiku | ₹96.30 | ₹481.50 | Anthropic | | Gemini 1.5 Pro | ₹120.38 | ₹481.50 | Google | | Claude 3.5 Sonnet | ~₹274 | ~₹1,370 | Anthropic | | Grok 4 | ~₹274 | ~₹1,370 | xAI | | Mistral Large | ₹192.60 | ₹577.80 | Mistral |

Reasoning models — cost per million tokens

Reasoning models generate hidden "thinking" tokens billed at the output rate, so real per-request cost is usually higher than the table implies — a single response can use thousands of reasoning tokens.

| Model | Input (₹/1M) | Output (₹/1M) | Provider | |-------|--------------|----------------|----------| | DeepSeek R1 | ₹52.92 | ₹211.00 | DeepSeek | | o3-mini | ~₹100 | ~₹402 | OpenAI | | o3 | ~₹183 | ~₹731 | OpenAI | | o1 | ~₹1,370 | ~₹5,481 | OpenAI |

Embeddings — cost per million tokens

Embeddings are input-only (no output tokens), which makes them cheap even at high volume — the real cost driver in a RAG pipeline is usually re-embedding the same documents repeatedly.

| Model | Price (₹/1M tokens) | Provider | |-------|----------------------|----------| | text-embedding-3-small | ~₹1.83 | OpenAI | | text-embedding-3-large | ~₹11.88 | OpenAI |

Audio and image — cost per unit

These are priced per unit, not per token, so they don't fit the tables above.

| Model | Unit | Approx. price | |-------|------|----------------| | TTS (tts-1) | 1,000 characters | ~₹1.37 | | Whisper transcription | 1 minute of audio | ~₹0.55 | | DALL-E 3 (1024×1024, standard) | 1 image | ~₹3.65 | | DALL-E 2 (1024×1024) | 1 image | ~₹1.46 |

How to read this table

Input vs. output matters more than the headline number. A model with cheap input but expensive output can cost more than expected on chat-style workloads where responses run long. Always multiply by your actual token counts, not the per-million rate alone.

Cache-friendly workloads change the math. If you repeatedly send the same system prompt or document, prompt caching cuts the effective input cost by up to 90% on supported models — see the prompt caching guide for the exact multipliers.

These are Standard-tier prices. Most providers also offer Batch tiers (roughly half price, asynchronous) for non-latency-sensitive workloads. AICredits surfaces these where available — check aicredits.in/models for the full tier breakdown per model.

Frequently Asked Questions

Why do these prices differ from the provider's official USD price?

They don't — they're the same USD price converted to INR using AICredits' forex rate plus a 5% buffer, so international transaction fees and rate volatility are priced in upfront rather than surprising you on a monthly statement.

How often does this list get updated?

Provider prices change periodically and the forex rate moves daily. This post is refreshed on a regular cadence — for the live number at any moment, use aicredits.in/calculator or aicredits.in/models.

Which model gives the best value for high-volume production use?

For most production chat workloads, Gemini 2.0 Flash, DeepSeek V3, and GPT-4o-mini sit in the best cost-to-capability range. Reserve the expensive models (Claude Sonnet, Grok 4, o1) for tasks that specifically need their extra reasoning or writing quality.

Get started

• Estimate your own workload against live rates before committing budget.
• Create a free account and top up via UPI to start testing models against real prompts.
• Browse the full, always-current list at aicredits.in/models.
• Related: How to Estimate Your LLM API Costs Before You Ship.