LLM API Cost Calculator: GPT vs Claude vs Gemini vs DeepSeek
Enter requests per day and tokens per request to see the monthly API cost on 17 models from OpenAI, Anthropic, Google, DeepSeek, xAI, Mistral, Qwen and Meta, with prompt caching and batch pricing.
Describe Your Workload
= 30,000 requests a month (30 days)
System prompt + chat history + retrieved documents + the user's message.
Count too: they bill as output.
Rule of thumb for English: is roughly 750 words.
With , the repeated start of each prompt is billed at the cheaper cache-read rate.
GPT-6.1 Sol
- Input: 45M tokens @ $2/M
- $90.00
- Output: 12M tokens @ $10/M
- $120.00
- Total per month
- $210.00
Cost = tokens ÷ 1,000,000 × price per 1M, summed for input, cached input and output
Cheapest mid-tier model for this workload: Mistral Large 4 at $56.70/month (73% less). Cheapest overall: DeepSeek V4.1 Flash at $8.03/month. Cheaper models can be weaker, so test them on your own prompts before switching.
This Workload on Every Model
45M input and 12M output tokens a month at real-time prices. Cheapest first; tap a model to load it above.
| Model | Tier | Input / 1M | Cached / 1M | Output / 1M | Context | Monthly cost | vs GPT-6.1 Sol |
|---|---|---|---|---|---|---|---|
| DeepSeek | Budget | $0.045 | $0.018 | $0.5 | 1M | $8.03 | -96% |
| Anthropic | Budget | $0.1 | $0.01 | $0.5 | 1M | $10.50 | -95% |
| OpenAI | Budget | $0.1 | $0.01 | $0.5 | 1.1M | $10.50 | -95% |
| Alibaba Qwen | Budget | $0.15 | $0.016 | $0.47 | 1M | $12.39 | -94% |
| Meta (open weights) | Budget | $0.1875 | $0.05 | $0.6525 | 524.3K | $16.27 | -92% |
| Budget | $0.25 | $0.025 | $1.5 | 1M | $29.25 | -86% | |
| Mistral | Mid | $0.7 | $0.07 | $2.1 | 1M | $56.70 | -73% |
| DeepSeek | Mid | $0.115 | $0.11 | $4.6 | 1M | $60.37 | -71% |
| Mid | $0.825 | $0.0825 | $4.125 | 1M | $86.63 | -59% | |
| OpenAI | Mid | $1 | $0.05 | $5 | 1.1M | $105.00 | -50% |
| xAI | Flagship | $2 | $0.5 | $6 | 500K | $162.00 | -23% |
| Anthropic | Mid | $2 | $0.1 | $10 | 1M | $210.00 | 0% |
| Flagship | $2 | $0.1 | $10 | 1M | $210.00 | 0% | |
| OpenAI | Mid | $2 | $0.1 | $10 | 1.1M | $210.00 | selected |
| Anthropic | Flagship | $4 | $0.2 | $20 | 1M | $420.00 | +100% |
| Anthropic | Flagship | $10 | $0.25 | $50 | 1M | $1,050.00 | +400% |
| OpenAI | Flagship | $10 | $1 | $50 | 1.1M | $1,050.00 | +400% |
Prices are the per-token rates OpenRouter listed on 4 Nov 2026. Buying direct from a provider can be priced differently, and some providers charge more for very long prompts, so check the provider's pricing page before you commit a budget. Cache-write charges (paid the first time a prefix is cached) are not included; with a high cache share they add little. A month is 30 days.
Frequently Asked Questions & How It Works
Quick answers on token pricing, prompt caching and batch discounts. Prices quoted were checked on OpenRouter on 4 November 2026.
How do I calculate LLM API cost?
Multiply the tokens you send and receive by the model's price per million tokens, then add the two. Example: 1,000 requests a day for 30 days is 30,000 requests. At 1,500 input and 400 output tokens each, that is 45M input and 12M output tokens. On GPT-6.1 Sol ($2 input / $10 output per 1M) the bill is 45 × $2 + 12 × $10 = $210 a month, or $7 per 1,000 requests.
Which LLM API is cheapest?
For that same 1,000-requests-a-day workload, the cheapest of the 17 models listed was DeepSeek V4.1 Flash at about $8 a month, followed by Claude Haiku 5.5 and GPT-6 Luna at $10.50 each. The most expensive, Claude Fable 5.1 and GPT-6 Astra, came to $1,050. Price alone doesn't pick a model: run a few dozen of your real prompts through the two or three cheapest that look good enough before you switch.
How much does prompt caching save?
It cuts the price of the repeated part of the prompt, not the output. On Claude Sonnet 5.5, cached input costs $0.10 per 1M tokens against $2 for normal input. If 80% of the 1,500 input tokens in the example above come from cache, the monthly bill falls from $210 to about $142. The saving is biggest for long system prompts, RAG context and agents that resend the same tools on every call.
Is the Batch API worth it?
Yes, for work that can wait. On OpenRouter, the OpenAI, Anthropic and Google models in this calculator list batch prices at half their real-time price or less, and Grok 4.7's batch rate is 20% lower. Results come back later rather than instantly, so use batch for evaluations, tagging, bulk summaries and back-fills, and keep chat and agents on the real-time API.
Why do output tokens cost more than input tokens?
Providers price generation higher than reading. For most models listed here, output costs 3 to 6 times as much as input, and on DeepSeek V4 Pro it is 40 times. So the length of the answer often matters more than the length of the prompt. Reasoning models also bill their hidden thinking tokens as output, which can multiply the output count.
Are these the same prices as buying direct from OpenAI or Anthropic?
Not always. The calculator uses OpenRouter's listed per-token rates because they come from one public, machine-readable list that can be re-checked automatically. Direct pricing can differ, and long-prompt surcharges, regional pricing, volume discounts and cache-write fees are not included. Check the provider's own pricing page before you commit a budget.
Building an AI Chatbot, Agent or RAG App?
Vsurge Media builds production AI assistants, RAG search and automation for businesses, and helps keep the monthly API bill under control.