LLM API cost calculator
Estimate what an AI API workload costs per month. Enter input and output tokens per request, the share of input served from the prompt cache, and requests per month: the calculator prices that workload on 40 generally available models from 8 providers and ranks them cheapest first, with the cost per request and per month. Use it as an OpenAI, Claude (Anthropic) or Gemini API cost calculator, or compare all three on the same numbers. List prices, USD, no signup.
Estimate your monthly AI API cost
Turn on the Batch API for asynchronous work, or include the 27 legacy and preview models. All math runs in your browser — nothing is sent anywhere. Not sure of your token counts? Check the OpenAI usage dashboard or Claude Console usage.
Cheapest for this workload: Qwen3.7-Flash at $56.00/mo ($0.0006/request)
Ranking uses current headline rates — intro and off-peak prices (flagged) may change.
| # | Model | Provider | Cost / request | Monthly cost | Applied |
|---|---|---|---|---|---|
| 1 | Qwen3.7-Flash | Alibaba (Qwen) | $0.0006 | $56.00 | |
| 2 | Ministral 3 3B | Mistral | $0.0012 | $120.00 | |
| 3 | Qwen-Flash | Alibaba (Qwen) | $0.0013 | $130.00 | |
| 4 | Gemini 2.5 Flash-Lite | $0.0018 | $180.00 | ||
| 5 | Ministral 3 8B | Mistral | $0.0018 | $180.00 | |
| 6 | Qwen3.5-Flash | Alibaba (Qwen) | $0.0018 | $180.00 | |
| 7 | Claude Haiku 5.5 | Anthropic | $0.0020 | $200.00 | |
| 8 | GPT-6 Luna | OpenAI | $0.0020 | $200.00 | |
| 9 | Ministral 3 14B | Mistral | $0.0024 | $240.00 | |
| 10 | Qwen3.8-Flash | Alibaba (Qwen) | $0.0024 | $244.00 | |
| 11 | DeepSeek V4.1 Flashoff-peak | DeepSeek | $0.0027 | $270.00 | |
| 12 | Mistral Small 4 | Mistral | $0.0027 | $270.00 | |
| 13 | GPT-5.6 Luna | OpenAI | $0.0044 | $440.00 | |
| 14 | Qwen-Plus | Alibaba (Qwen) | $0.0064 | $640.00 | |
| 15 | Qwen3.7-Plus | Alibaba (Qwen) | $0.0072 | $720.00 | |
| 16 | Gemini 2.5 Flash | $0.0080 | $800.00 | ||
| 17 | Gemini 3.5 Flash-Lite | $0.0080 | $800.00 | ||
| 18 | DeepSeek V4 Prooff-peak | DeepSeek | $0.0106 | $1,056.00 | |
| 19 | Gemini 3.8 Flashintro | $0.0150 | $1,500.00 | ||
| 20 | GPT-5.4 mini | OpenAI | $0.0165 | $1,650.00 | |
| 21 | Grok 4.3 | xAI | $0.0175 | $1,750.00 | |
| 22 | Claude Haiku 4.5 | Anthropic | $0.0200 | $2,000.00 | |
| 23 | GPT-6.1 Terra | OpenAI | $0.0200 | $2,000.00 | |
| 24 | Muse Spark 1.3 | Meta | $0.0210 | $2,100.00 | |
| 25 | Grok 4.7 | xAI | $0.0320 | $3,200.00 | |
| 26 | Qwen3.8-Max | Alibaba (Qwen) | $0.0320 | $3,200.00 | |
| 27 | Gemini 2.5 Pro | $0.0325 | $3,250.00 | ||
| 28 | Claude Sonnet 5.5 | Anthropic | $0.0400 | $4,000.00 | |
| 29 | GPT-6.1 Sol | OpenAI | $0.0400 | $4,000.00 | |
| 30 | GPT-6 Sol | OpenAI | $0.0400 | $4,000.00 | |
| 31 | Gemini 4 Argonintro | $0.0400 | $4,000.00 | ||
| 32 | GPT-5.6 Terra | OpenAI | $0.0440 | $4,400.00 | |
| 33 | GPT-5.4 | OpenAI | $0.0550 | $5,500.00 | |
| 34 | Claude Opus 5.5 | Anthropic | $0.0800 | $8,000.00 | |
| 35 | GPT-5.6 Solintro | OpenAI | $0.0800 | $8,000.00 | |
| 36 | GPT-5.5 | OpenAI | $0.1100 | $11,000.00 | |
| 37 | Claude Fable 5.1 | Anthropic | $0.2000 | $20,000.00 | |
| 38 | Claude Mythos 5.1 | Anthropic | $0.2000 | $20,000.00 | |
| 39 | GPT-6 Astra | OpenAI | $0.2000 | $20,000.00 | |
| 40 | GPT-5.5 Pro | OpenAI | $0.6600 | $66,000.00 |
List prices, USD, directional. Rates are provider list prices per 1M tokens and are meant for comparison, not billing. Batch rates shown as 50% off are derived where a provider offers batch but does not publish a separate figure. Preview, promo, intro, peak/off-peak, long-context, and third-party-host prices are labeled where they apply. Token counts vary by tokenizer, so per-token price is not always a like-for-like cost. Always confirm with the provider before relying on a number. Prices as of 2026-11-02.
How the estimate is calculated
- The math. Cost per request = (input × (1 − cached share) × input rate + input × cached share × cached-input rate + output × output rate) ÷ 1,000,000, with rates in USD per 1M tokens. Monthly cost = cost per request × requests per month. Models are ranked by monthly cost, cheapest first, and each links to its full price matrix.
- Prompt caching. The cached share of input is billed at the provider's cached-input rate. A model with no published cached rate bills all input at the input rate and is marked "no cache".
- Batch API. With batch on, input and output use the batch rate for Anthropic, OpenAI, Google and Mistral: the published batch rate, or 50% of the standard rate where the provider offers batch but publishes no separate figure. Other providers keep standard rates and are marked "no batch". Cached tokens stay at the standard cached-input rate, so where a provider also discounts cached tokens in batch, the real bill can be a little lower.
- Flags. "intro" means the model is ranked at an introductory rate that is scheduled to rise; "off-peak" means it is ranked at the off-peak rate and peak hours cost more.
- Not included. Cache-write charges, long-context surcharges, peak-hour rates, reasoning tokens you don't count as output, tool and search fees, taxes, regional pricing, negotiated discounts, and the rates Amazon Bedrock, Vertex AI and Azure charge for the same models. Image, video and embedding models are out of scope.
- Sources. Rates come from each provider's official pricing page — the same committed dataset the LLM pricing comparison and model pages show. Token counts vary by tokenizer, so the same text can cost different amounts on two models with the same per-token price.
FAQ
How do I calculate LLM API cost?
Multiply each kind of token by its rate. Per request: input tokens × (1 − cached share) × input rate + input tokens × cached share × cached-input rate + output tokens × output rate, with rates in USD per 1M tokens, divided by 1,000,000. Multiply by requests per month for the monthly cost. This calculator does that for every model at once and ranks them from cheapest to most expensive.
Can I use it as an OpenAI, Claude or Gemini API cost calculator?
Yes. It prices 18 OpenAI models, 11 Anthropic Claude models and 11 Google Gemini API models, plus xAI Grok, DeepSeek, Mistral, Meta and Alibaba Qwen — 67 models from 8 providers. Every model is priced on the same workload, so you can compare a GPT model and a Claude model side by side. Only generally available models are ranked until you include legacy and preview ones.
How do prompt caching and the Batch API change the estimate?
The cached share of input is billed at the provider's cached-input rate instead of the input rate; a model with no published cached rate bills all input at the input rate and is marked "no cache". With Batch API on, input and output are billed at the batch rate for providers that offer one (Anthropic, OpenAI, Google and Mistral) — the published batch rate, or 50% of the standard rate where no separate figure is published. Other models keep standard rates and are marked "no batch". Cached tokens stay at the standard cached-input rate with batch on.
Does it cover Amazon Bedrock, Google Vertex AI or Azure OpenAI?
No. It uses each provider's own API list prices; Google rows are Gemini API prices. The same models bought through Amazon Bedrock, Vertex AI or Azure are billed at those platforms' own published rates, which can differ, so check the cloud provider's pricing page. The AI API price calendar lists announced model shutdown dates on Amazon Bedrock and Azure.
Are reasoning tokens included?
Only if you count them. Reasoning models generate hidden reasoning tokens that are billed as output, so include them in your output tokens per request. Check your provider's usage dashboard for real counts before trusting a per-request figure.
Is this what I will actually be billed?
No. It is a directional estimate from provider list prices in USD, as of 2026-11-02. It does not include cache-write charges, long-context surcharges, peak-hour rates, tool or search fees, taxes, regional pricing or negotiated discounts. Introductory and off-peak rates are flagged in the results because they can change. Confirm with the provider before relying on a number.
Do you see the numbers I enter?
No. The calculation runs in your browser over prices built into the page, and nothing you type is sent anywhere.
From an estimate to what you actually spend
This calculator prices a workload you describe, at list prices. Your real bill depends on the tokens your apps actually send and the models they actually call. CloudQuell tracks what you actually spend: your OpenAI and Anthropic usage on one ledger next to your AWS, Azure and GCP bills, with anomaly alerts when a line jumps.
Track what you actually spend →Related: LLM pricing comparison →AI API price calendar →LLM pricing glossary →