AI API cost calculator (₹)

AI API cost calculator — rupee-based budget planning

Building with LLMs? Predict monthly spend accurately by calculating token costs per model, then compare. unoblox's cost calculator shows real ₹ pricing across all major models — no hidden fees, transparent costs.

How token costs work

Every API call consumes two types of tokens:

• Input tokens: Your prompt (typically 200–2000 per request).
• Output tokens: The model's response (typically 100–1000 per request).

Cost = (Input tokens × input rate) + (Output tokens × output rate).

Rates vary by model. GPT-4o is ₹253.07/M input, ₹1,012.28/M output. Llama Scout is ₹10.12/M input, ₹30.36/M output. Same 1M tokens, 25× price difference.

Step-by-step cost calculation

Scenario: A customer-support chatbot handling 100K daily customer questions.

1. Estimate daily tokens:

   • Input per question: 300 tokens (context + instructions + customer message).
   • Output per answer: 150 tokens (model response).
   • Daily volume: 100K questions × 300 input + 100K × 150 output = 45M daily tokens.
   • Monthly: 45M × 30 = 1.35B monthly tokens.

2. Pick a model and calculate cost** (900M input + 450M output — the 2:1 split implied by the 300:150 ratio above):

   • Qwen3.8-27B (₹0/M): ₹0 / month ✓
   • Llama Scout (₹10.12 input, ₹30.36 output): 900 × ₹10.12 + 450 × ₹30.36 = ₹9,108 + ₹13,662 = ₹22,770 / month.
   • Claude Sonnet (₹202.46 input, ₹1,012.28 output): 900 × ₹202.46 + 450 × ₹1,012.28 = ₹182,214 + ₹455,526 = ₹637,740 / month.
   • GPT-4o (₹253.07 input, ₹1,012.28 output): 900 × ₹253.07 + 450 × ₹1,012.28 = ₹227,763 + ₹455,526 = ₹683,289 / month.

Takeaway: The same workload costs ₹0 (free tier), ~₹23K (Llama Scout), or ~₹683K (GPT-4o). Model choice is often more important than optimization.

Model pricing reference (per 1M tokens)

Model	Input (₹)	Output (₹)	₹/monthly (1B input, 350M output)
Qwen3.8-27B	0	0	₹0
Qwen3 235B	9.11	55.67	₹28,595
Llama Scout	10.12	30.36	₹20,746
Claude Sonnet	202.46	1,012.28	₹556,758
Claude Opus	506.14	2,530.71	₹1,391,889
GPT-4o	253.07	1,012.28	₹607,368
GPT-4.1	202.46	809.82	₹485,897
GPT-5	126.53	1,012.28	₹480,828

Monthly budget by use case

Use case	Est. monthly tokens	Recommended model	₹/month
Chatbot (1K daily users)	50M	Llama Scout	₹843
RAG pipeline (100K docs)	2B	Qwen3 235B	₹49,260
Batch summarization	500M	Qwen3.8-27B	₹0
Production SaaS	5B	Llama 4 Scout	₹84,333
Enterprise reasoning	1B	Claude Sonnet	₹472,400

Assumes a 2:1 input:output token mix, the same ratio used throughout this page.

How to reduce costs

1. Optimize prompts: Shorter, clearer instructions = fewer input tokens.

   • Before: "You are a customer support bot. Please respond to the following query…" (25 tokens).
   • After: "Support bot response:" (3 tokens). 8× token savings.

2. Use cheaper models first: Start on Qwen3.8-27B or Llama Scout. Upgrade only if output quality is insufficient.

3. Batch requests: Process 1000 questions at once instead of 1000 individual API calls — same token cost, lower latency overhead.

4. Cache context: If your app repeats the same 10K-token context (e.g., a company handbook), cache it and reuse.

5. Streaming: Only pay for tokens you generate. Stop early if the output is complete.

Using the unoblox calculator

Visit https://unoblox.ai/pricing to:

• Enter your estimated monthly tokens.
• Select 2–3 models to compare.
• See total ₹/month in real-time.
• Export a CSV for your finance team.

Frequently asked questions

How do I know if my token estimate is accurate? Call https://api.unoblox.ai/v1/tokenize (demo) or enable usage logging in your SDK. Most apps underestimate tokens by 20–30%.

Do I get charged per API call or per token? Per token. 10 calls with 1M tokens each = same cost as 1 call with 10M tokens.

Can I set a monthly spend cap? Yes. unoblox allows max_spend_monthly: ₹5000 per API key. Requests over the limit return 429 Limit Exceeded.

What about caching or context reuse? Where a model page lists a Cached input price (for example DeepSeek, OpenAI GPT and Gemini models), the prompt tokens the provider serves from its cache are billed at that lower rate. Claude models on unoblox do not offer cached-input pricing yet, so every Claude prompt token is billed at the input rate. Check the model page before relying on caching.

Should I optimize for input or output tokens? Input is typically 3–5× more output. Optimize input first (prompt engineering). Output optimization happens naturally (better models = shorter answers).

How often does pricing change? unoblox posts 30-day notice for price changes. Current rates valid through Q4 2026.

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