Which AI model should you use?
Paste a prompt. See the model that gives the most for the money, and what it costs.
Task
Compare side by side2 or more models, effort, your plan
Optionsfiles, what the model must do, makers and hosts
The model must be able to:
Every model, cheapest first
Quality ratings for this taskfrom Arena votes
How this worksin five lines
Cost is your prompt and answer in tokens times each model's listed price. A token is about 4 characters of English, so costs are estimates.
Best value is the cheapest model whose Arena rating cannot be told apart from the top model's: people compared their answers and voted.
The task is guessed from your words, in your browser. Your prompt is never sent or saved.
The badge is the chance that model's price changes in the next 90 days, from the history of every change in the catalogue.
Effort is how hard a thinking model works on your prompt. The suggestion is a rule of thumb from the task and your words; the levels each model takes come from LiteLLM.
Find a model
Every model in LiteLLM's catalogue, with its price, its history and the chance its price changes.
About the names
The maker and the grouped names come from naming rules (for
example, eu.anthropic.claude-3-5-sonnet-20241022-v2:0 and
snowflake/claude-3-5-sonnet are both grouped as anthropic /
claude-3.5-sonnet). The host is LiteLLM's own provider field.
How long does a price last?
Of the 5,383 models added to LiteLLM's catalogue since 6 September 2023, these shares had their first change within a year.
First change, by time since a model was added
Hover over a bar to see its value. Removals include bulk clean ups of the catalogue.
The numbers, with 95% confidence intervals
| Within | Price change | Retirement announced | Removed | None of these | Still followed |
|---|---|---|---|---|---|
| 30 days | 5.9%5.2 to 6.5 | 3.5%3.0 to 4.0 | 10.0%9.2 to 10.8 | 80.7%79.6 to 81.7 | 3,788 |
| 90 days | 8.5%7.7 to 9.3 | 5.2%4.5 to 5.8 | 13.0%12.1 to 13.9 | 73.3%72.1 to 74.5 | 2,717 |
| 6 months | 10.8%9.9 to 11.8 | 7.7%6.9 to 8.5 | 13.9%12.9 to 14.8 | 67.6%66.3 to 68.9 | 2,341 |
| 1 year | 14.7%13.6 to 15.8 | 16.5%15.2 to 17.7 | 16.1%15.0 to 17.2 | 52.8%51.2 to 54.4 | 1,215 |
| 2 years | 18.3%16.9 to 19.7 | 29.8%27.9 to 31.7 | 23.1%21.6 to 24.7 | 28.8%26.8 to 30.9 | 379 |
Aalen-Johansen estimates of competing risks with 1,000 bootstrap resamples. Each row adds up to 100% because the exits compete: once a model is removed, its price can no longer change. Download as CSV.
Does the definition matter?
| What counts as a price change | Within 1 year |
|---|---|
| Any price field (main result) | 14.7%13.6 to 15.8 |
| Input and output token prices only | 12.5%11.5 to 13.6 |
| Ignoring changes in the first week | 12.1%11.0 to 13.2 |
| Leaving out corrections | 12.3%11.2 to 13.5 |
Every change, or real changes only?
Real changes only leaves out unit fixes (a price made 100 times larger or smaller) and price changes undone within 30 days, and treats a model removed in a bulk clean up (50 or more removed on the same day) as still followed until that day.
| Within 1 year | Price change | Retirement announced | Removed | None of these |
|---|---|---|---|---|
| Every change (main result) | 14.7%13.6 to 15.8 | 16.5%15.2 to 17.7 | 16.1%15.0 to 17.2 | 52.8%51.2 to 54.4 |
| Real changes only | 12.3%11.2 to 13.5 | 18.7%17.2 to 20.2 | 5.7%4.9 to 6.5 | 63.3%61.5 to 64.9 |
Changed in the last 7 days
Tags mark bulk clean ups (50 or more models removed on the same day), unit fixes (a price made 100 times larger or smaller) and reverted changes (undone within 30 days).
Most repriced
Changes from one price to another; first prices, unit fixes and changes undone within 30 days are not counted.
Hosts most likely to change a price within a year
Only hosts with at least 100 models still followed at one year.
About
llmledger reads the full git history of LiteLLM's model price file every day, and the Arena leaderboard for quality. It estimates what a prompt costs, which model is the best value, and how likely each price is to change.
How often is the automatic task guess right?
What the estimates leave out
Prompt tokens are counted at about 4 characters per token; each model counts its own way, often 10 to 30% differently. Thinking models also bill hidden reasoning as answer tokens. Caching, batch discounts and long prompt prices are left out. When LiteLLM has no per-image or per-second price, an image counts as about 1,000 tokens, a PDF page about 1,500, a second of audio about 32 and a second of video about 300. "How many times" is how often you send the prompt; in a chat, each message also re-sends the conversation so far.
Compare side by side: higher effort makes a thinking model write more hidden answer tokens, and no open source measures how many at each level yet, so the table starts every model at the answer length you picked; type the tokens you see in your own use to compare. On a Claude or ChatGPT plan you pay no per-token price; the table assumes a plan is used up in proportion to each model's API price, because plans do not publish per-model limits.
Data, licences and method
Data: LiteLLM main at f81e657, last price file
change 16 October 2026, 13:52 UTC. Page built 16 October 2026, 17:39 UTC with Python 3.14.7.
Method, specification and tests: llmledger on GitHub. MIT licence.