AI API pricing compared
What it costs to call today's large language models through their APIs, across Amazon, Anthropic, Cohere, DeepSeek, Google, Mistral, OpenAI and xAI.
At a glance
The input list shows every model that charges under $0.20 per million input tokens, in three price bands. The frontier list shows each provider's most capable model. At the extremes, for a million input tokens plus a million output tokens, the most expensive models cost around 343 times as much as the cheapest.
Input under $0.20
Bars show input and output on a $0 to $0.60 scale, so you can see how much output adds.
-
Under $0.10 input · 3 models
-
Nova MicroAmazon$0.035 in · $0.14 out
- $0.0375 in · $0.15 out
-
Nova LiteAmazon$0.06 in · $0.24 out
-
-
$0.10 to $0.15 input · 4 models
-
Ministral 3 3BMistral$0.10 in · $0.10 out
-
Gemini 2.5 Flash-LiteGoogle$0.10 in · $0.40 out
-
gpt-6-lunaOpenAI$0.10 in · $0.50 out
-
Claude Haiku 5.5 (prompt ≤ 100k)Anthropic$0.10 in · $0.50 out
-
-
$0.15 to $0.20 input · 4 models
-
Ministral 3 8BMistral$0.15 in · $0.15 out
-
Z.ai GLM 5.3 Flash (hosted)Mistral$0.15 in · $0.50 out
-
Mistral Small 4Mistral$0.15 in · $0.60 out
- $0.15 in · $0.60 out
-
Frontier models
Each provider's most capable model, cheapest output first. Bars show input and output on the price table's $0 to $50 scale.
-
Mistral Large 4Mistral$0.70 in · $2.10 out
- $1.32 in · $3.96 out
-
grok-4.7xAI$2.00 in · $6.00 out
- $2.00 in · $10.00 out
- $2.50 in · $10.00 out
-
Nova PremierAmazon$2.50 in · $12.50 out
-
Claude Fable 5.2 / Mythos 5.2*Anthropic$8.00 in · $40.00 out
-
gpt-6-astraOpenAI$10.00 in · $50.00 out
How AI API pricing works
AI providers charge per token, a chunk of text roughly three-quarters of a word long. Prices below are in US dollars per million tokens, split into three rates:
- Input
- The text you send to the model: your prompt, instructions and any documents.
- Cached input
- A discounted rate for input the provider has already processed recently, such as a repeated system prompt.
- Output
- The text the model writes back. This is usually the most expensive part.
What 1,000 requests cost
Per-token prices only tell half the story: what you pay depends on how much text your application sends and gets back. Set the sliders to match a typical request. Input cost runs left of the centre line and output cost to the right, so you can see which side drives each model's total.
Under $1 15 models
- Nova Micro
- Command R7B
- Ministral 3 3B
- Nova Lite
- Ministral 3 8B
- Ministral 3 14B
- Gemini 2.5 Flash-Lite
- gpt-6-luna
- Claude Haiku 5.5 (prompt ≤ 100k)
- Z.ai GLM 5.3 Flash (hosted)
- Mistral Small 4
- Command R (08-2024)
- Codestral (25.08)
- gpt-5.6-luna
- DeepSeek V4.1 Flash
$1 to $5 17 models
- Gemini 3.1 Flash-Lite
- Gemini 2.5 Flash
- Gemini 3.5 Flash-Lite
- Nova 2 Lite
- Mistral Large 4
- Claude Haiku 5.5 (prompt > 100k)
- grok-build-0.1
- Gemini 3 Flash Preview
- Nova Pro
- grok-4.3
- grok-4.20-0309 (reasoning / non-reasoning / multi-agent)
- Gemini 3.8 Flash
- Gemini 3.6 Flash
- DeepSeek V4 Pro
- Claude Haiku 4.5
- gpt-6.1-terra
- Z.ai GLM 5.3 (hosted)
$5 to $20 21 models
- grok-4.5
- grok-4.6
- grok-4.7
- Gemini 2.5 Pro (prompt ≤ 200k)
- Nova 2 Pro
- gpt-6.1-sol
- Claude Sonnet 5.5
- gpt-6-sol
- Claude Sonnet 5
- Gemini 4 Argon
- Command A
- Gemini 3.1 Pro Preview (prompt ≤ 200k)
- gpt-5.6-terra
- Nova Premier
- gpt-5.3-codex
- Gemini 2.5 Pro (prompt > 200k)
- Claude Sonnet 4.6 / 4.5
- Gemini 3.1 Pro Preview (prompt > 200k)
- Claude Opus 5.5
- gpt-5.6-sol
- Claude Opus 5 / 4.8 / 4.7 / 4.6 / 4.5
$20 and over 5 models
- chat-latest
- Claude Fable 5.2 / Mythos 5.2*
- gpt-6-astra
- Claude Fable 5.1 / Mythos 5.1*
- Claude Fable 5 / Mythos 5*
At 1,000 input and 500 output tokens per request, with no cached input. Models with no listed cached price pay their full input price for cached tokens.
Price comparison
| Input Output | ||||||
|---|---|---|---|---|---|---|
| Mistral | Ministral 3 3B | 256k | $0.10 | $0.01 | $0.10 | |
| Amazon | Nova Micro | 128k | $0.035 | $0.00875 | $0.14 | |
| Mistral | Ministral 3 8B | 256k | $0.15 | $0.015 | $0.15 | |
| Cohere | Command R7B Check current pricing with Cohere | 128k | $0.0375 | — | $0.15 | |
| Mistral | Ministral 3 14B | 256k | $0.20 | $0.02 | $0.20 | |
| Amazon | Nova Lite | 300k | $0.06 | $0.015 | $0.24 | |
| Gemini 2.5 Flash-Lite | — | $0.10 | $0.01 | $0.40 | ||
| OpenAI | gpt-6-luna | — | $0.10 | $0.01 | $0.50 | |
| Anthropic | Claude Haiku 5.5 (prompt ≤ 100k) | — | $0.10 | $0.01 | $0.50 | |
| Mistral | Z.ai GLM 5.3 Flash (hosted) | 1M | $0.15 | $0.03 | $0.50 | |
| Mistral | Mistral Small 4 | 256k | $0.15 | $0.015 | $0.60 | |
| Cohere | Command R (08-2024) Check current pricing with Cohere | 128k | $0.15 | — | $0.60 | |
| Mistral | Codestral (25.08) | 128k | $0.30 | $0.03 | $0.90 | |
| DeepSeek | DeepSeek V4.1 Flash Half price outside peak hours | 1M | $0.30 | $0.006 | $1.20 | |
| OpenAI | gpt-5.6-luna | — | $0.20 | $0.02 | $1.20 | |
| Gemini 3.1 Flash-Lite | — | $0.25 | $0.025 | $1.50 | ||
| xAI | grok-build-0.1 | 256k | $1.00 | $0.20 | $2.00 | |
| Mistral | Mistral Large 4 | 1M | $0.70 | $0.07 | $2.10 | |
| Anthropic | Claude Haiku 5.5 (prompt > 100k) | — | $0.50 | $0.05 | $2.50 | |
| xAI | grok-4.3 | 1M | $1.25 | $0.20 | $2.50 | |
| xAI | grok-4.20-0309 (reasoning / non-reasoning / multi-agent) | 1M | $1.25 | $0.20 | $2.50 | |
| Gemini 2.5 Flash Deprecated: moves to Gemini 3.5 Flash-Lite on 1 December 2026 | 1M | $0.30 | $0.03 | $2.50 | ||
| Gemini 3.5 Flash-Lite | — | $0.30 | $0.03 | $2.50 | ||
| Amazon | Nova 2 Lite | 1M | $0.33 | $0.0825 | $2.75 | |
| Gemini 3 Flash Preview | — | $0.50 | $0.05 | $3.00 | ||
| Amazon | Nova Pro | 300k | $0.80 | $0.20 | $3.20 | |
| Gemini 3.8 Flash Rises to $1.50 / $0.15 / $7.50 on 1 January 2027 | — | $0.75 | $0.075 | $3.75 | ||
| Gemini 3.6 Flash Rises to $1.50 / $0.15 / $7.50 on 1 January 2027 | — | $0.75 | $0.075 | $3.75 | ||
| DeepSeek | DeepSeek V4 Pro Half price outside peak hours | 1M | $1.32 | $0.044 | $3.96 | |
| Mistral | Z.ai GLM 5.3 (hosted) | 1M | $1.40 | $0.14 | $4.40 | |
| Anthropic | Claude Haiku 4.5 | — | $1.00 | $0.10 | $5.00 | |
| OpenAI | gpt-6.1-terra | — | $1.00 | $0.05 | $5.00 | |
| xAI | grok-4.5 | 500k | $2.00 | $0.30 | $6.00 | |
| xAI | grok-4.6 | 500k | $2.00 | $0.50 | $6.00 | |
| xAI | grok-4.7 | 500k | $2.00 | $0.50 | $6.00 | |
| OpenAI | gpt-6.1-sol | — | $2.00 | $0.10 | $10.00 | |
| Anthropic | Claude Sonnet 5.5 | — | $2.00 | $0.10 | $10.00 | |
| OpenAI | gpt-6-sol | — | $2.00 | $0.20 | $10.00 | |
| Anthropic | Claude Sonnet 5 | — | $2.00 | $0.20 | $10.00 | |
| Gemini 2.5 Pro (prompt ≤ 200k) Deprecated: moves to Gemini 3.8 Flash on 1 December 2026 | — | $1.25 | $0.125 | $10.00 | ||
| Gemini 4 Argon Rises to $4.00 / $0.20 / $20.00 on 1 December 2026 | — | $2.00 | $0.10 | $10.00 | ||
| Cohere | Command A Check current pricing with Cohere | 256k | $2.50 | — | $10.00 | |
| Amazon | Nova 2 Pro | — | $1.375 | — | $11.00 | |
| Gemini 3.1 Pro Preview (prompt ≤ 200k) | — | $2.00 | $0.20 | $12.00 | ||
| OpenAI | gpt-5.6-terra | — | $2.00 | $0.20 | $12.00 | |
| Amazon | Nova Premier | 1M | $2.50 | $0.625 | $12.50 | |
| OpenAI | gpt-5.3-codex | — | $1.75 | $0.175 | $14.00 | |
| Anthropic | Claude Sonnet 4.6 / 4.5 | — | $3.00 | $0.30 | $15.00 | |
| Gemini 2.5 Pro (prompt > 200k) Deprecated: moves to Gemini 3.8 Flash on 1 December 2026 | — | $2.50 | $0.25 | $15.00 | ||
| Gemini 3.1 Pro Preview (prompt > 200k) | — | $4.00 | $0.40 | $18.00 | ||
| Anthropic | Claude Opus 5.5 | — | $4.00 | $0.20 | $20.00 | |
| OpenAI | gpt-5.6-sol Promotional price until 30 November 2026 | — | $4.00 | $0.40 | $20.00 | |
| Anthropic | Claude Opus 5 / 4.8 / 4.7 / 4.6 / 4.5 | — | $5.00 | $0.50 | $25.00 | |
| OpenAI | chat-latest | — | $5.00 | $0.50 | $30.00 | |
| Anthropic | Claude Fable 5.2 / Mythos 5.2* | — | $8.00 | $0.20 | $40.00 | |
| OpenAI | gpt-6-astra | — | $10.00 | $1.00 | $50.00 | |
| Anthropic | Claude Fable 5.1 / Mythos 5.1* | — | $10.00 | $0.25 | $50.00 | |
| Anthropic | Claude Fable 5 / Mythos 5* | — | $10.00 | $1.00 | $50.00 |
Prices in USD per million tokens. “—” means the provider doesn't list a figure. The bars draw every model's input and output prices on one scale. Click a column heading to sort, or a model name to open its page on the provider's site. Gemini prices are Google's Standard paid tier for text input, from Google's pricing page (checked 10 November 2026); Gemini context caching also charges an hourly storage fee. *Claude Mythos models have limited availability and are open only to approved organisations; they cost the same as the matching Claude Fable model. DeepSeek prices are its standard (peak) rates from DeepSeek's pricing page; they halve outside its peak hours of 01:00–04:00 and 06:00–10:00 UTC on weekdays (noon–3 pm and 5–9 pm in Sydney during daylight saving). Amazon Nova prices are on-demand rates in AWS's US East (N. Virginia) region, from the AWS price list (2 November 2026); Nova Micro, Lite and Pro cost about 5% more in Sydney, Nova 2 Lite reaches Australia only through AWS's global routing, and Nova Premier runs only in the US. xAI prices are for prompts under 200k tokens; they double for a request whose prompt reaches 200k. Cohere no longer publishes per-token prices for these models on its public pages, so the figures shown may be out of date: check current pricing with Cohere before relying on them. Its newest models, Command A+ and Command A Reasoning, have no published per-token price, so they aren't listed.
The cheapest models are small, fast economy models built for simple, high-volume work such as classifying, extracting and routing text. Most providers now offer one at or near the lowest price point, but their output prices vary much more, so compare both rates against how much text your application sends and receives.
Frontier models are the top tier from their providers, designed for the most demanding reasoning, coding and multi-step tasks. Their prices vary widely, from $2.10 to $50.00 per million output tokens. For most business applications, a mid-range model offers a better balance of cost and capability.
Why older models can cost more
Providers often increase the prices of their older economy models to encourage developers to move to their current models. You can see the pattern in the table: Claude Haiku 4.5 costs ten times as much as Claude Haiku 5.5 (for prompts up to 100k tokens), and older Claude Sonnet and Opus versions cost more than the latest releases. Older models are also retired over time, so if you build on an AI API, review your model choice regularly. Moving to a current model is often both cheaper and more capable.
Where your data is processed
Price isn't the only thing to compare. DeepSeek is among the cheapest providers here, but its own API processes and stores data in China, and the Australian Government banned DeepSeek from its systems and devices in 2025. DeepSeek's models are open-weight, so cloud providers such as AWS and Azure also host them, with data kept in the region you choose, at their own prices. Amazon's Nova Micro, Lite and Pro models run in AWS's Sydney region. If your data must stay in Australia, check where each provider processes it before you choose a model.