Set your traffic and prompt sizes. 17 models are priced side by side, with long-context surcharges, batch discounts and the specs that decide whether a model fits.
108M tokens a month. Claude Haiku 5.5 is the cheapest option; GPT-6 Astra costs 100× more.
Send simple requests to a cheap model and only the hard ones to a strong model. Uses your traffic, cache and batch settings above.
Hard requests are often longer, especially with reasoning models. Set their size above, then check answer quality on your own data before you switch.
List prices per million tokens, standard rate, before batch or caching discounts.
| Model | Providers | Input $/1M | Cached $/1M | Output $/1M | Context |
|---|---|---|---|---|---|
| Claude Fable 5.1 | Anthropic | $10.00 | $0.25 | $50.00 | 1M |
| Claude Opus 5.5 | Anthropic | $4.00 | $0.20 | $20.00 | 1M |
| Claude Sonnet 5.5 | Anthropic | $2.00 | $0.10 | $10.00 | 1M |
| Claude Haiku 5.5 | Anthropic | $0.10 | $0.01 | $0.50 | 1M |
| DeepSeek V4 Pro | DeepSeek | $0.66 | $0.02 | $1.98 | 1M |
| DeepSeek V4.1 Flash | DeepSeek | $0.15 | $0.0030 | $0.60 | 1M |
| Gemini 4 Argon | $2.00 | $0.10 | $10.00 | 1.05M | |
| Gemini 3.1 Pro | $2.00 | $0.20 | $12.00 | 1.05M | |
| Gemini 3.8 Flash | $0.75 | $0.08 | $3.75 | 1.05M | |
| Gemini 3.5 Flash-Lite | $0.30 | $0.03 | $2.50 | 1.05M | |
| Llama 4 Maverick | Meta | $0.19 | $0.05 | $0.65 | 128K |
| GPT-6 Astra | OpenAI | $10.00 | $1.00 | $50.00 | 1.05M |
| GPT-6.1 Sol | OpenAI | $2.00 | $0.10 | $10.00 | 1.05M |
| GPT-6.1 Terra | OpenAI | $1.00 | $0.05 | $5.00 | 1.05M |
| GPT-6 Luna | OpenAI | $0.10 | $0.01 | $0.50 | 1.05M |
| Qwen3.8-27B | Qwen | $0.15 | $0.04 | $1.88 | 262K |
| Grok 4.7 | xAI | $2.00 | $0.50 | $6.00 | 500K |
? = not confirmed by a published source yet.
A token is a piece of text the model reads or writes. In English, 1,000 tokens is about 750 words. Most other languages need more tokens for the same text, and Arabic often needs about twice as many as English. Each provider has its own tokenizer, so counts differ a little. The text estimator above gives a rough figure.
Writing a reply takes more computing than reading a prompt, so providers charge several times more per output token. Long answers and reasoning models raise the bill fast, so the output setting often matters more than it looks.
When many requests start with the same text, such as a system prompt, a policy or a document, the provider can keep it and bill later reads at a lower rate. Set the cached share to the part of your input that repeats. Some providers also charge to write or store the cache; those fees are not included here.
When nobody waits for the answer: tagging, overnight summaries, data extraction. Results arrive within up to 24 hours at about half price. It does not suit chatbots or anything interactive.
From each provider's official pricing page, linked at the bottom of this page. Prices for open models come from third-party hosts, named next to the model. The date at the top shows when prices were last checked. Check the official page before you commit to a budget.
Not always. Price says nothing about answer quality on your task. Shortlist two or three models here, then test them on 10 to 20 of your real examples. The Which model? pages list candidates by task.