Compare AI model prices and see your monthly bill
API prices of 393 language models from OpenAI, Anthropic, Google and 62 other providers. Say how much you use, and the table sorts every model by what you would pay.
- Free, no sign-up
- No API key needed
- Prices collected 3 Oct 2026
Every model, sorted by your monthly bill
Prices are per million tokens (a token is about three quarters of a word). “Your cost” applies the usage above to each model. Click a column to sort, a model for its details.
157 of 393 models, sorted by your monthly cost — loading the complete list…
Each row: input / output price, then your monthly cost. Token prices per 1M tokens. Tick up to 4 models to compare.
“Your cost” applies the cache-read price only where the source lists one; a model with no price data shows a dash and is never ranked as cheapest.
Collected from the OpenRouter models API (466 entries). 99 models are cross-checked against provider list prices in the LiteLLM price table.
Prices by provider
Open a provider to see all of its models and what each would cost you.
- OpenAI65 modelsfrom $0.018 per 1M input
- Qwen54 modelsfrom $0.03 per 1M input
- Google30 modelsfrom $0.05 per 1M input
- Mistral19 modelsfrom $0.019 per 1M input
- Z.ai16 modelsfrom $0.06 per 1M input
- Anthropic15 modelsfrom $1.00 per 1M input
- DeepSeek14 modelsfrom $0.017 per 1M input
- NVIDIA11 modelsfrom $0.05 per 1M input
- Meta8 modelsfrom $0.027 per 1M input
- MiniMax8 modelsfrom $0.20 per 1M input
- MoonshotAI7 modelsfrom $0.45 per 1M input
- Tencent7 modelsfrom $0.044 per 1M input
- xAI7 modelsfrom $1.00 per 1M input
How to read the table
- Describe your usagePick a ready-made scenario or type how many requests you make and how long they are.
- We do the sum for every modelYour volume is multiplied by each model’s published prices. Nothing is estimated.
- Read the rankingThe table sorts by your monthly bill. Open a model for every price and its source.
Language-model APIs charge per token, with one price for what you send (input) and a higher one for what the model writes (output). Prices are quoted per million tokens. A request that sends 3,000 tokens and receives 600 uses 0.003 and 0.0006 of those units, which is why a single call costs a fraction of a cent and why the monthly volume is what matters.
Input, output and cached input
The Input and Output columns are the prices OpenRouter charges. Cached in is the lower price some models charge to re-read a prefix they have already processed; a dash means none is published. The thin bars under each price compare magnitudes on a logarithmic scale, because prices in this table span more than three orders of magnitude.
Your cost per month
The last column applies the workload at the top of the page to each row: requests per day, times 30 days, times tokens per request, times the model’s prices. Pick a scenario in plain language or type your own four numbers. The calculator shows the same sum line by line for up to four models.
What the table does not tell you
Price says nothing about whether a model is good at your task. Where OpenRouter relays the Intelligence Index from Artificial Analysis you can sort by it, but the only test that counts is running your own prompts. Reasoning models also bill their hidden thinking as output, so their effective output volume is larger than the visible answer.
Guides: prompt caching and batch discounts, long-context pricing, cheapest models by category.
Why use this table
Ranked by your bill, not by a headline price
The default order multiplies your input and output volumes by each model’s prices. Change the workload and the order changes.
Two sources, shown side by side
99 models carry both the OpenRouter price and the provider list price; 8 of them differ by more than 1% and are flagged.
Nothing estimated
Cached-input prices are the published ones (236 models list one). Where a source is silent the table shows a dash, not a guess.
Dated, and flagged when old
The collection time is printed under the table, and a warning appears when it is more than 36 hours old.
Free and unpriced models kept apart
The 22 free models have their own filter, and the 7 entries without price data are never ranked as cheapest.
A link reproduces the screen
Workload, filters, sort order, currency and the models picked for comparison are all in the page address.
More price tools and guides
- Cost calculatorYour monthly bill on up to four models, line by line.
- Cheapest modelsThe five lowest-cost models in each category.
- Free modelsModels listed at zero price, kept apart from the paid ones.
- Price changesA dated log of every price that went up or down.
- Caching and batch discountsHow the two discounts work and what you save.
- Long-context pricingModels that read very long prompts, and what they charge.
Questions and answers
Where do these prices come from?
The catalogue and the price of each model come from the public OpenRouter models API. For models sold directly by Anthropic, OpenAI, Google, DeepSeek, xAI and Mistral, the provider list price comes from the LiteLLM price table, which records the address of the provider’s pricing page. When the two differ by more than 1% the row says so.
Why is the table ordered by monthly cost and not by price per token?
Because the cheapest input price is rarely the cheapest bill. A model with cheap input and expensive output loses to another one as soon as your requests produce long answers. The table multiplies your own input and output volumes by each model’s prices, so the order changes when your workload changes.
How often are the prices updated?
A scheduled job collects the catalogue once a day. The date and time of the last collection are printed above the table, and a warning appears when that collection is more than 36 hours old. The pages you are reading were built from the collection of 3 Oct 2026, 08:40 UTC.
Is the OpenRouter price the same as buying from the provider?
Often, but not always. OpenRouter routes each model through one or more hosts and charges its own price. Open a model to see the provider list price next to the OpenRouter price, with the batch price and the price above 200K tokens where the provider publishes them.
How is the cached-input price used in the cost?
If you say that part of your input is read from cache, that part is billed at the cache-read price the source lists for the model. If the model lists no cache-read price, the cached part is billed as normal input: no discount is assumed. The cost of writing to the cache is not included.
Why do some models show a dash instead of a price?
A dash means the source gives no price for that field, as happens with routers that pick a model for you. It is different from a free model, whose price is zero. Models without a price are kept at the end of the table and are never ranked as the cheapest.
Can I see the prices in Brazilian reais?
Yes. Choose BRL and every price and cost is converted with the PTAX selling rate published by the Banco Central do Brasil. The rate and its date are shown next to the table. Your card or bank will apply a different rate and may add taxes.