LLM API cost calculator
What your workload costs on every Claude, GPT and Gemini model, at live prices. Change the numbers; the table re-sorts by cost.
| Model | Per request | Per day | Per 30 days |
|---|---|---|---|
| Claude Haiku 5.5 Anthropic | $0.00032 | $0.315 | $9.45 |
| GPT-5.6 Luna OpenAI | $0.00071 | $0.71 | $21.30 |
| Gemini 3 Flash Google | $0.0018 | $1.78 | $53.25 |
| Gemini 3.8 Flash Google | $0.0024 | $2.36 | $70.88 |
| Gemini 3.6 Flash Google | $0.0024 | $2.36 | $70.88 |
| Claude Haiku 4.5 Anthropic | $0.0032 | $3.15 | $94.50 |
| Gemini 3.5 Flash Google | $0.0053 | $5.33 | $159.75 |
| Claude Sonnet 5.5 Anthropic | $0.0062 | $6.15 | $184.50 |
| GPT-6.1 Sol OpenAI | $0.0062 | $6.15 | $184.50 |
| Claude Sonnet 5 Anthropic | $0.0063 | $6.30 | $189.00 |
| Gemini 4 Argon Google | $0.0063 | $6.30 | $189.00 |
| GPT-6 Sol OpenAI | $0.0063 | $6.30 | $189.00 |
| GPT-5.6 Sol OpenAI | $0.0063 | $6.30 | $189.00 |
| Gemini 3.1 Pro Google | $0.0071 | $7.10 | $213.00 |
| GPT-5.6 Terra OpenAI | $0.0071 | $7.10 | $213.00 |
| Claude Sonnet 4.6 Anthropic | $0.0095 | $9.45 | $283.50 |
| Claude Sonnet 4.5 Anthropic | $0.0095 | $9.45 | $283.50 |
| Claude Opus 5.5 Anthropic | $0.012 | $12.30 | $369.00 |
| Claude Opus 5 Anthropic | $0.016 | $15.75 | $472.50 |
| Claude Opus 4.8 Anthropic | $0.016 | $15.75 | $472.50 |
| Claude Opus 4.7 Anthropic | $0.016 | $15.75 | $472.50 |
| Claude Opus 4.6 Anthropic | $0.016 | $15.75 | $472.50 |
| Claude Opus 4.5 Anthropic | $0.016 | $15.75 | $472.50 |
| Claude Fable 5.1 Anthropic | $0.03 | $30.38 | $911.25 |
| Claude Fable 5 Anthropic | $0.032 | $31.50 | $945.00 |
| GPT-6 Astra OpenAI | $0.032 | $31.50 | $945.00 |
Prices per the lowest-priced channel group; cache writes are not included. Model marketplace · Cut Claude API costs with prompt caching
Frequently asked questions
How is the cost of an LLM API request calculated?
Input tokens × the input price + output tokens × the output price, each per million tokens. Cached input tokens are billed at the cache-read price instead, and writing a cache at the cache-write price.
Are reasoning (thinking) tokens included?
Yes. Reasoning tokens are billed as output tokens, so count them in the output tokens per request.
How much does prompt caching save?
Cached input is billed at the cache-read price, a fraction of the normal input price, so a long system prompt or document reused on every call costs much less. Set the cache share to see the difference.
Where do these prices come from?
From the live catalog: the same prices the API charges and the marketplace shows, at the lowest-priced channel group for each model.