AI token cost calculator
Work out what an LLM feature will cost before you ship it. Enter tokens per request and volume, or paste a real prompt, and compare 33 current models side by side, including prompt caching and batch discounts.
| Model | Input /1M | Output /1M | Per request | Per month |
|---|---|---|---|---|
| GPT-5 nanoOpenAI | $0.05 | $0.40 | $0.000235 | $7.05 |
| GPT-6 LunaOpenAI | $0.10 | $0.50 | $0.00035 | $10.50 |
| Claude Haiku 5.5Anthropic | $0.10 | $0.50 | $0.000455 | $13.65 |
| GPT-5.6 LunaOpenAI | $0.20 | $1.20 | $0.00078 | $23.40 |
| GPT-5.4 nanoOpenAI | $0.20 | $1.25 | $0.0008 | $24.00 |
| DeepSeek V4.1 FlashDeepSeek | $0.30 | $1.20 | $0.00093 | $27.90 |
| Gemini 3.1 Flash-LiteGoogle | $0.25 | $1.50 | $0.000975 | $29.25 |
| GPT-5 miniOpenAI | $0.25 | $2.00 | $0.00118 | $35.25 |
| Gemini 3.5 Flash-LiteGoogle | $0.30 | $2.50 | $0.00145 | $43.50 |
| Grok Build 0.1xAI | $1.00 | $2.00 | $0.0023 | $69.00 |
| Gemini 3.8 FlashGoogle | $0.75 | $3.75 | $0.00262 | $78.75 |
| Grok 4.3xAI | $1.25 | $2.50 | $0.00287 | $86.25 |
| GPT-5.4 miniOpenAI | $0.75 | $4.50 | $0.00292 | $87.75 |
| Claude Haiku 4.5Anthropic | $1.00 | $5.00 | $0.0035 | $105.00 |
| DeepSeek V4 ProDeepSeek | $1.32 | $3.96 | $0.00356 | $106.92 |
| Grok 4.5xAI | $2.00 | $6.00 | $0.0054 | $162.00 |
| Grok 4.7xAI | $2.00 | $6.00 | $0.0054 | $162.00 |
| GPT-5OpenAI | $1.25 | $10.00 | $0.00588 | $176.25 |
| Gemini 4 ArgonGoogle | $2.00 | $10.00 | $0.007 | $210.00 |
| GPT-6 SolOpenAI | $2.00 | $10.00 | $0.007 | $210.00 |
| GPT-6.1 SolOpenAI | $2.00 | $10.00 | $0.007 | $210.00 |
| Gemini 3.1 Pro (preview)Google | $2.00 | $12.00 | $0.0078 | $234.00 |
| GPT-5.6 TerraOpenAI | $2.00 | $12.00 | $0.0078 | $234.00 |
| Claude Sonnet 5Anthropic | $2.00 | $10.00 | $0.0091 | $273.00 |
| Claude Sonnet 5.5Anthropic | $2.00 | $10.00 | $0.0091 | $273.00 |
| GPT-5.4OpenAI | $2.50 | $15.00 | $0.00975 | $292.50 |
| GPT-5.6 SolOpenAI | $4.00 | $20.00 | $0.014 | $420.00 |
| Claude Opus 5.5Anthropic | $4.00 | $20.00 | $0.0182 | $546.00 |
| GPT-5.5OpenAI | $5.00 | $30.00 | $0.0195 | $585.00 |
| Claude Opus 5Anthropic | $5.00 | $25.00 | $0.0227 | $682.50 |
| GPT-6 AstraOpenAI | $10.00 | $50.00 | $0.035 | $1,050 |
| Claude Fable 5Anthropic | $10.00 | $50.00 | $0.0455 | $1,365 |
| Claude Fable 5.1Anthropic | $10.00 | $50.00 | $0.0455 | $1,365 |
30,000 requests a month · prices per 1M tokens, USD · month = 30 days
How to read the numbers
Input vs output
Output tokens cost 2–8× more than input on most models. Long answers and reasoning ("thinking") tokens usually dominate the bill, so estimate output carefully.
Caching is the biggest lever
A long system prompt or document reused across calls can be read from cache at 75–99% off. Agents and chat apps with stable prefixes often cache 70–90% of input.
Long prompts can cost more
Several models switch to a higher rate for very long prompts (over 272K tokens on newer GPT models, 200K on Gemini Pro and Grok). The table flags when that happens.
Cheapest current models for a chatbot reply
1,500 input / 400 output tokens, standard pricing.
- GPT-5 nano (OpenAI): $0.235 per 1,000 replies
- GPT-6 Luna (OpenAI): $0.35 per 1,000 replies
- Claude Haiku 5.5 (Anthropic): $0.455 per 1,000 replies
- GPT-5.6 Luna (OpenAI): $0.78 per 1,000 replies
- GPT-5.4 nano (OpenAI): $0.80 per 1,000 replies
Frequently asked questions
How is the cost calculated?
Cost per request = (uncached input tokens × input price + cached input tokens × cached price + output tokens × output price) ÷ 1,000,000. Monthly cost multiplies that by your request volume, with a month counted as 30 days. If a prompt is longer than a model's long-context threshold (for example 272K tokens on newer OpenAI models or 200K on Gemini Pro and Grok), the higher long-context rate is applied to the whole request, as the providers do.
How current are these prices?
Every price was checked by hand against the provider's official pricing page on October 19, 2026. Each model page links to its source. Announced price changes (such as Gemini 4 Argon's increase on December 1, 2026) are flagged in the table.
Do output tokens include reasoning or "thinking" tokens?
Yes. Reasoning models bill their hidden thinking tokens as output tokens, so a short visible answer can still use thousands of output tokens. If you use a reasoning model, raise the output figure to include the thinking budget you expect.
What is cached input?
When the start of your prompt (system message, tool definitions, a long document) repeats across requests, providers can serve it from a cache at a large discount, typically 75–99% off the normal input price. Set the cached share to the part of each prompt that stays the same. Anthropic also charges extra to write to the cache (1.25× input for a 5-minute cache), which this calculator does not include.
Why does Claude count more tokens for the same text?
Anthropic says Claude 4.7 and later models use a newer tokenizer that produces about 30% more tokens for the same text. With the tokenizer adjustment on, token counts for those models are multiplied by 1.3 so that costs are compared for the same piece of text.
Is the token count exact?
Counts use OpenAI's open-source o200k_base tokenizer, the encoding used by GPT-4o, GPT-4.1, GPT-5 and the o-series. Other providers use their own tokenizers, so counts for their models are estimates that are usually within about 10–20% for English text and can differ more for code and non-English text. For billing-grade numbers, use the provider's own token-counting endpoint.
Is the text I paste uploaded anywhere?
No. Token counting runs entirely in your browser. The site has no backend and does not store or send what you paste.