How to read the table
- Input and output. Input is everything you send (instructions, history, documents); output is what the model writes. On the models in the table, output costs 5× to 6× input.
- Caching. A prefix that repeats across calls can be read from cache, at 2.5% to 10% of the input price. Writing to the cache costs more than normal input, and each model has a minimum size for a prefix to be cached.
- Batch. On the Batch API, a request can take up to 24 hours and input and output cost half.
- Long prompts. Some models bill the whole call at a higher price when input goes over a limit; the rule appears under the model name.
Price per 1 million tokens
Tap a column title to sort. The initial order is by provider and input price.
| Claude Haiku 5.5 above 100,000 tokens: 5× input, 5× output | Anthropic | $0.10 | $0.50 | $0.0100 | $0.125 1 h: $0.20 | $0.0500 | $0.25 | 1,000,000 | 128,000 | $0.45 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | $0.10 | $1.25 1 h: $2.00 | $0.50 | $2.50 | 200,000 | 64,000 | $4.50 |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $10.00 | $0.10 | $2.50 1 h: $4.00 | $1.00 | $5.00 | 1,000,000 | 128,000 | $9.00 |
| Claude Opus 5.5 | Anthropic | $4.00 | $20.00 | $0.20 | $5.00 1 h: $8.00 | $2.00 | $10.00 | 1,000,000 | 128,000 | $18.00 |
| Claude Fable 5.1 | Anthropic | $10.00 | $50.00 | $0.25 | $12.50 1 h: $20.00 | $5.00 | $25.00 | 1,000,000 | 128,000 | $45.00 |
| GPT-6 Luna above 272,000 tokens: 2× input, 1.5× output | OpenAI | $0.10 | $0.50 | $0.0100 | $0.125 | $0.0500 | $0.25 | 1,050,000 | 128,000 | $0.45 |
| GPT-5.6 Luna above 272,000 tokens: 2× input, 1.5× output | OpenAI | $0.20 | $1.20 | $0.0200 | $0.25 | $0.10 | $0.60 | 1,050,000 | 128,000 | $1.00 |
| GPT-6 Sol above 272,000 tokens: 2× input, 1.5× output | OpenAI | $2.00 | $10.00 | $0.20 | $2.50 | $1.00 | $5.00 | 1,050,000 | 128,000 | $9.00 |
| GPT-6.1 Sol above 272,000 tokens: 2× input, 1.5× output | OpenAI | $2.00 | $10.00 | $0.10 | $2.50 | $1.00 | $5.00 | 1,050,000 | 128,000 | $9.00 |
| GPT-5.6 Terra above 272,000 tokens: 2× input, 1.5× output | OpenAI | $2.00 | $12.00 | $0.20 | $2.50 | $1.00 | $6.00 | 1,050,000 | 128,000 | $10.00 |
| GPT-5.6 Sol promotional price at least through November 21, 2026 above 272,000 tokens: 2× input, 1.5× output | OpenAI | $4.00 | $20.00 | $0.40 | $5.00 | $2.00 | $10.00 | 1,050,000 | 128,000 | $18.00 |
| GPT-6 Astra above 272,000 tokens: 2× input, 1.5× output | OpenAI | $10.00 | $50.00 | $1.00 | $12.50 | $5.00 | $25.00 | 1,050,000 | 128,000 | $45.00 |
Prices before taxes. The "1,000 calls" column uses 2,000 input and 500 output tokens per call.
Worked example: 1,000 calls
Each call sends 2,000 tokens and gets 500 back. The math is (input tokens × input price + output tokens × output price) ÷ 1,000,000:
- Claude Haiku 5.5: (2,000 × $0.10 + 500 × $0.50) ÷ 1,000,000 = $0.000450 per call; × 1,000 = $0.45.
- GPT-6 Astra: (2,000 × $10.00 + 500 × $50.00) ÷ 1,000,000 = $0.0450 per call; × 1,000 = $45.00.
Between the cheapest and the most expensive model in the table, the same task costs 100× more. With 90% of input read from cache, the 1,000 calls on GPT-6.1 Sol would drop to $5.58; on the Batch API, to $4.50.
Sources and date
Prices copied from each provider's official pages into the site's price file on October 14, 2026:
- Anthropic: platform.claude.com/docs/en/about-claude/pricing
- OpenAI: developers.openai.com/api/docs/pricing
The table is built from that file: when a price changes there, this page changes with it.
To do your own math
Each model has a page with the detailed price and usage math: click its name in the table. For your volume, use the API cost calculator, the price comparison or the prompt caching calculator.