AI model pricing, side by side.
Price per million input and output tokens, context window, and provider. Filter, sort, and see which run on open weights, you can also host those yourself.
- Module
- Router
- Input
- Nothing, it is already ready
- Runs on
- Your browser
- Cost
- Free, no signup
| Model | Input / 1M | Output / 1M | Context |
|---|---|---|---|
| Nova LiteAmazon | 0.06 | 0.24 | 300k |
| Gemini 2.5 Flash-LiteGoogle | 0.10 | 0.40 | 1,000k |
| Qwen 3 32BAlibaba · open weights | 0.10 | 0.30 | 128k |
| Llama 4 ScoutMeta · open weights | 0.11 | 0.34 | 10,000k |
| Mistral SmallMistral · open weights | 0.20 | 0.60 | 128k |
| DeepSeek V3DeepSeek · open weights | 0.27 | 1.10 | 128k |
| Llama 4 MaverickMeta · open weights | 0.27 | 0.85 | 1,000k |
| Gemini 3 FlashGoogle | 0.30 | 2.50 | 1,000k |
| Grok 4 minixAI | 0.30 | 0.50 | 128k |
| GPT-4.1 miniOpenAI | 0.40 | 1.60 | 1,000k |
| DeepSeek R1DeepSeek · open weights | 0.55 | 2.19 | 128k |
| GPT-5 miniOpenAI | 0.60 | 2.40 | 400k |
| Nova ProAmazon | 0.80 | 3.20 | 300k |
| Claude Haiku 4.5Anthropic | 1.00 | 5.00 | 200k |
| o4-miniOpenAI | 1.10 | 4.40 | 200k |
| Qwen 3 MaxAlibaba | 1.20 | 6.00 | 262k |
| GPT-4.1OpenAI | 2.00 | 8.00 | 1,000k |
| Mistral LargeMistral | 2.00 | 6.00 | 128k |
| Gemini 3 ProGoogle | 2.50 | 15.00 | 2,000k |
| Command ACohere · open weights | 2.50 | 10.00 | 256k |
| Claude Sonnet 5Anthropic | 3.00 | 15.00 | 1,000k |
| Grok 4xAI | 3.00 | 15.00 | 256k |
| GPT-5OpenAI | 10.00 | 30.00 | 400k |
| Claude Opus 5Anthropic | 15.00 | 75.00 | 1,000k |
Figures in US dollars per million tokens, public list price. Updated 2026-08-11.
Price is not a footnote: it is an architecture decision
Between the most expensive and the cheapest model in this table, there is a difference of more than one hundred times per token. This is not a market variation, it is the difference between a product that works financially and one that does not. Those who choose a model at the start of a project and never review it again are usually paying for capacity they do not use.
Beyond price, two columns matter: the context window, which defines how much fits in one call, and the open-weight label. An open-weight model can be hosted by you, which changes the conversation from "how much does it cost to rent" to "how much does it cost to own".
- 01
Pick a provider, or see them all
The table starts with the whole market and filters by who supplies it.
- 02
Sort by whatever decides it in your case
Lowest input if you send a lot of text; lowest output if you generate a lot; largest context if you work with long documents.
- 03
Notice which ones ship open weights
You can run these on your own infrastructure, and that is where price becomes sovereignty.
- 01
What does price per million tokens mean?
- It is the market's standard unit: how much it costs to process one million word pieces. One million tokens is roughly equivalent to about 750 thousand words in Portuguese, around ten average-length books.
- 02
Why are input and output billed separately?
- Because they cost different things. Reading input is a single pass; generating output is token by token, and uses much more computing. That is why output usually costs three to five times more than input, and why ignoring it when estimating cost gets the calculation half wrong.
- 03
What are "open weights" and why does it matter?
- It is a model whose weights have been published: you can download and run it on your own machine or server. It matters because it changes the nature of the relationship, instead of renting access to a box that can change price or disappear, you gain the option to operate it yourself.
- 04
Is a large context window worth paying more for?
- Only if you use it. Context is how much fits in one call, not how much the model "knows". Paying for a window of millions of tokens to send two-thousand-word prompts is buying space that stays empty.
- 05
How often do these prices change?
- Often, there were cuts of more than 90% in some families throughout 2024 and 2025, and new models arrive every month. That is why the update date is visible in the table footer: a price number without a date is a number you cannot trust.
- 06
Can I use several models without keeping several accounts?
- You can, with a Router in front: one key, and the decision about which model handles each call stays on the inside. That is what the Router module does, and what keeps you from being locked into whoever raised the price.
You picked a model. And when the price changes?
The Router module puts a single key in front of the whole market and picks the destination per task. Switching models stops being a migration and becomes a setting.