Models (LLMs)

At the top of the prompt page you can find two dropdowns to select an inference API provider and an LLM from their portfolio to complete your prompt.

You can find an up-to-date list of all supported models in our LLM Index.

Token price

The inference price for input and output tokens is set by each provider and displayed to the right of the model selection in units of cents per 1,000 tokens. There is no markup, inference costs will be the same as if you would use the provider's API directly.

The total completion price is always calculated as

$$\text{completion cost}=\frac{n_{input}}{1000}pric{e}_{input}+\frac{n_{output}}{1000}pric{e}_{output}$$

Model settings

Below the model selection you can tune the model settings. To edit a parameter, just click on it to open its details. The popover lets you adjust the value of the parameter and gives you useful information about the min/max range, the default value, and an explanation of what the parameter does.

If a specific model does not support a setting it will be grayed out and omitted when submitting the prompt to the API.

Token limit

The maximum number of tokens the model should process/generate in the completion. Unfortunately, different providers treat the token limit differently. Sometimes the token limit refers to input plus output tokens, e.g. OpenAI, and sometimes it refers only to the output tokens, e.g. Anthropic.

The model can never exceed the token limit. So if you experience completion cutoffs, the token limit is most likely the culprit.

Temperature

The sampling temperature to use. Higher values like 0.9 will make the output more random, while lower values like 0.2 will make it more focused and deterministic.

Top p

An alternative to sampling with temperature, called nucleus sampling, where the model considers the results of the tokens with top_p probability mass. So 0.1 means only the tokens comprising the top 10% probability mass are considered.

It is generally recommended to either alter the temperature or the top p parameter, but not both at the same time.

Frequency penalty

Positive values penalize new tokens based on their existing frequency in the text so far, decreasing the model's likelihood to repeat the same line verbatim.

Presence penalty

Positive values penalize new tokens based on whether they appear in the text so far, increasing the model's likelihood to talk about new topics.

Seed

If specified, the model will make a best effort to sample deterministically, such that repeated requests with the same seed and parameters should return the same result. However, determinism is not guaranteed!

JSON mode

If set to true, the model is forced to output JSON-formatted responses. Note that your prompt has to contain the word "json", otherwise the API will throw an error.

Stop sequences

A comma-separated list of up to 4 sequences where the API will stop generating further tokens. The returned text will not contain the stop sequence.

Related resources

• How to choose the right LLM for your use case (blog post)