AI API pricing comparison (₹)

Every major model, one invoice, rupees

unoblox bills all models on a single monthly GST invoice in rupees. No hidden fees, no international gateway tax. Compare live rates below — all indicative; see each model page for the current price.

Prices differ by model because compute cost differs: a frontier reasoning model like Claude Opus runs on far more GPU per token than a lean model like DeepSeek V4 Flash or Qwen3 235B. unoblox passes through the real provider-side rate plus a platform fee — no markup games, no bundling.

2026 pricing snapshot (per 1M tokens)

Model	Input ₹	Output ₹	Best for	Speed
DeepSeek V4 Flash	9.07	18.14	Budget tasks, speed	<1s
Qwen3-1.7B	0	0	Free prototyping	<500ms
Qwen3 235B-A22B	9.07	55.44	Code, vision, cheap	2–3s
Kimi K2.7	68.5	342.7	Reasoning, logic	3–5s
Claude Sonnet	201.6	1008	Nuanced, safe	2–3s
GPT-4o	252	1008	Vision, multi-modal	<2s
GPT-5	126	1008	Long-form reasoning	5–10s
Claude Opus	504	2520	Hardest reasoning	10–15s
Llama 4 Maverick	20.16	80.64	Open-weight reasoning	3–5s
Gemma 4 31B	13.10	38.30	Open-weight, cheap	1–2s

How to save 40%+ on AI costs

1. Use the cheapest model that clears your quality bar. DeepSeek V4 Flash or Qwen3 235B handle most classification, extraction, and chat tasks; reserve Claude Opus or GPT-5 for the fraction of requests that truly need frontier reasoning.
2. Batch and reuse context. Grouping requests and reusing a shared system prompt cuts overhead compared to many small, one-off calls.
3. Cache prompts. Reuse the same long context (a knowledge base, a system prompt) across calls instead of resending it every time — this cuts both cost and latency on repeated contexts (RAG, multi-turn chat).
4. Free embedding + reranking. Qwen3-Embedding (₹0) + Qwen3-Reranker (₹0) means your RAG pipeline's retrieval step costs nothing — only the final generation call is billed.
5. Match context window to the task. Don't pay output-token prices for a model's extra reasoning if a shorter, cheaper model already gets the answer right on the first try.

Frequently asked questions

Q: Which model should I pick? Start with DeepSeek V4 Flash (₹9.07 input). Move to Claude Sonnet if you need nuance; GPT-4o if you need vision.

Q: Do these prices ever change? Yes — rates track upstream provider costs and can move. unoblox posts 30-day notice before any price increase, and current live rates always show on each model's /models page.

Q: Can I lock in a price? Yes. Commit to a monthly spend; volume discounts apply (ask in the admin dashboard).

Q: What about the free tier? ₹0 for Qwen3-1.7B, embeddings, reranking, and on sign-up credits (₹250/month initially). Upgrade as you scale.

Q: Do you have batch APIs like OpenAI? Not yet; coming Q4. For now, batch via client-side concurrency (1000 req/s limit per key).

Q: Can I claim GST input tax credit on my usage? Yes. Every invoice carries unoblox's GSTIN and a per-model line-item breakdown, so registered businesses can claim input tax credit the same way they would for any other B2B software spend.

Get started in rupees → https://unoblox.ai/sign-in