AI API prices compared: cost per 1 million tokens

Checked 26 October 2026 · by A.Frick

A table of AI API list prices per 1 million tokens, with source and check date for every row.

API prices are the base cost of building with AI. This table gathers them in one place so you can compare models side by side.

How to read the table

Unit. Every price is in US dollars per 1 million tokens (see price per million tokens). A token is a word piece, not a whole word.
Input. What you send: your prompt and any files or history.
Output. What the model writes back. This is usually the dearer part.
Cache. A lower price for input the provider has stored from an earlier call. See prompt caching.
No VAT. All prices are list prices without tax.
Date. Each row shows the day we checked it against the provider's page.

Provider

Provider	Model	Tier	Input	Output	Cached input	Context	Checked
Anthropic	Claude Fable 5.1 ¹	top	$10.00	$50.00	$0.25	1M	2026-10-26
Anthropic	Claude Opus 5.5 ²	top	$4.00	$20.00	$0.20	1M	2026-10-26
Anthropic	Claude Sonnet 5.5 ³	mid	$2.00	$10.00	$0.10	1M	2026-10-26
Anthropic	Claude Haiku 5.5 ⁴	small	$0.10	$0.50	$0.01	1M	2026-10-26
DeepSeek	DeepSeek V4 Pro ⁵	top	$1.32	$3.96	$0.044	1M	2026-10-26
DeepSeek	DeepSeek V4.1 Flash ⁶	small	$0.30	$1.20	$0.006	1M	2026-10-26
Google (Gemini API)	Gemini 4 Argon ⁷	top	$2.00	$10.00	$0.10	1.05M	2026-10-26
Google (Gemini API)	Gemini 3.8 Flash ⁸	mid	$0.75	$3.75	$0.075	1.05M	2026-10-26
Google (Gemini API)	Gemini 3.1 Flash-Lite ⁹	small	$0.25	$1.50	$0.025	1.05M	2026-10-26
Google (Gemini API)	Gemini 3.5 Flash-Lite ¹⁰	small	$0.30	$2.50	$0.03	1.05M	2026-10-26
Mistral	Mistral Large 4 ¹¹	top	$1.36	$4.18	$0.14	1M	2026-10-26
Mistral	Mistral Medium 3.5 ¹²	mid	$1.50	$7.50	–	256k	2026-10-26
Mistral	Ministral 3 8B ¹³	small	$0.15	$0.15	–	256k	2026-10-26
Mistral	Mistral Small 4 ¹⁴	small	$0.15	$0.60	–	256k	2026-10-26
Moonshot AI (Kimi)	Kimi K3 ¹⁵	top	$3.00	$15.00	$0.30	1.05M	2026-10-26
Moonshot AI (Kimi)	Kimi K3.1 Pro ¹⁶	top	$2.00	$10.00	$0.20	1.05M	2026-10-26
OpenAI	GPT-6 Astra ¹⁷	top	$10.00	$50.00	$1.00	1.05M	2026-10-26
OpenAI	GPT-6.1 Sol ¹⁸	mid	$2.00	$10.00	$0.10	1.05M	2026-10-26
OpenAI	GPT-6.1 Terra ¹⁹	mid	$1.00	$5.00	$0.05	1.05M	2026-10-26
OpenAI	GPT-6 Luna ²⁰	small	$0.10	$0.50	$0.01	1.05M	2026-10-26
xAI (Grok)	Grok 4.7 ²¹	top	$2.00	$6.00	$0.50	500k	2026-10-26
xAI (Grok)	Grok 4.3 ²²	mid	$1.25	$2.50	$0.20	1M	2026-10-26
xAI (Grok)	Grok Build 0.1 ²³	small	$1.00	$2.00	$0.20	256k	2026-10-26

USD per 1 million tokens, list price, excl. tax. Prices as of 26 October 2026.

Notes on individual models

Claude Fable 5.1: Cache read costs 2.5% of input price. Newer tokenizer: about 30% more tokens per text than Claude 4.6 and earlier.
Claude Opus 5.5: Anthropic's default pick for most workloads. Cache read costs 5% of input price. Newer tokenizer: about 30% more tokens per text than Claude 4.6 and earlier.
Claude Sonnet 5.5: Cache read costs 5% of input price. Newer tokenizer: about 30% more tokens per text than Claude 4.6 and earlier.
Claude Haiku 5.5: Tiered by prompt length: prompts over 100,000 tokens cost $0.50 input, $2.50 output, $0.05 cache read. Newer tokenizer: about 30% more tokens per text than Claude 4.6 and earlier.
DeepSeek V4 Pro: Peak-hour price (Mon-Fri 01:00-04:00 and 06:00-10:00 UTC). Off-peak is half: $0.66 input, $1.98 output, $0.022 cache. Version V4-Pro-0813. No vision. Open weights are on Hugging Face; the licence was not checked.
DeepSeek V4.1 Flash: Peak-hour price (Mon-Fri 01:00-04:00 and 06:00-10:00 UTC). Off-peak is half: $0.15 input, $0.60 output, $0.003 cache. API name deepseek-flash (version V4.1-Flash-1017). Open weights are on Hugging Face; the licence was not checked.
Gemini 4 Argon: Promotional price: $2 / $10 (cache $0.10) only through Nov 30, 2026; from Dec 1, 2026 $4 / $20 (cache $0.20). Max output 65,536 by default, up to 1,048,576 with long decode. Not on the free tier. Implicit caching only.
Gemini 3.8 Flash: Price rises from Jan 1, 2027 to $1.50 input, $7.50 output, $0.15 cache. Cache storage $0.50 per 1M tokens per hour (from 2027: $1.00).
Gemini 3.1 Flash-Lite: Price is for text, image, video input; audio input $0.50 (cache $0.05). Cache storage $1.00 per 1M tokens per hour.
Gemini 3.5 Flash-Lite: Same price for text, image, video and audio input. Cache storage $1.00 per 1M tokens per hour.
Mistral Large 4: Public preview since Oct 6, 2026. Priced below Medium 3.5. The mistral.ai FAQ still quotes $0.5 / $1.5 for 'Mistral Large'; the model page was used. Weights are open; the licence is not named on the page. Max output not stated.
Mistral Medium 3.5: Open weights under a Modified MIT licence. Costs more than Large 4. No cache price listed. Max output not stated.
Ministral 3 8B: Apache 2.0 licence. One price for input and output. No cache price listed. Ministral 3 3B costs $0.10.
Mistral Small 4: Apache 2.0 licence. No cache price listed. Max output not stated.
Kimi K3: Previous generation but dearer than K3.1 Pro. Thinking is always on. Max output is the API limit (default 131,072). Cache write $3 (5 min) or $6 (1 h). K2 models are being retired by end of 2026 and are not listed.
Kimi K3.1 Pro: Thinking is always on. Max output is the API limit (default 131,072). Cache write $2 (5 min) or $4 (1 h). Highspeed variant costs $4 / $20. Open weights announced for release by Nov 4, 2026, so not yet counted as open.
GPT-6 Astra: Prompts over 272K input tokens: $20 input, $2 cache, $75 output for the full request. Cache writes $12.50.
GPT-6.1 Sol: Prompts over 272K input tokens: $4 input, $0.20 cache, $15 output for the full request.
GPT-6.1 Terra: Prompts over 272K input tokens: $2 input, $0.10 cache, $7.50 output for the full request.
GPT-6 Luna: Prompts over 272K input tokens: $0.20 input, $0.02 cache, $0.75 output for the full request.
Grok 4.7: Prompts of 200k tokens or more: $4 input, $1 cache, $12 output for the full request. Max output not stated.
Grok 4.3: Prompts of 200k tokens or more: $2.50 input, $0.40 cache, $5 output for the full request. Max output not stated.
Grok Build 0.1: Cheapest Grok text model. Prompts of 200k tokens or more: $2 input, $0.40 cache, $4 output. Also known as grok-code-fast-1. No batch API. Max output not stated.

For where the numbers come from and how we check them, see the methodology.

What the table does not show

Discounts. Large buyers and some plans get lower prices.
Batch. The batch API often costs less in return for slower answers.
Tiers. Some models charge more above a certain prompt size.
Regional prices. Hosting in a given region can cost more.
Extras. Tools such as search, image input or long reasoning can add cost.
Tax and exchange. Your invoice may include VAT and your bank's rate.

To turn these prices into a monthly figure, use the cost calculator.

Questions and answers

What does 1 million tokens cost?

It depends on the model and on input or output. Small models can cost cents, top models several dollars per million. The table shows the current list price for each row.

Why does output cost more than input?

The model reads your input in one pass but writes its answer one token at a time, which ties up more compute. Providers therefore charge more per output token. The table lists both.

How many words are 1 million tokens?

In English roughly 750,000 words, since one token is about three quarters of a word. German and code use more tokens per word. Try the token estimator.

Are these the prices I will pay?

They are list prices in USD without VAT. Discounts, batch use, tiers and regional prices can change your bill. Check the provider's page before you commit.

Sources

Was this page helpful?

Ask a question, report an error or suggest a topic