Baseten

AISDK provides first-class support for Baseten with fully typed model APIs. Model capabilities are enforced at compile time using Rust's type system. This prevents model capability mismatches and guarantees the selected model is valid for the task (e.g. tool calling).

Installation

Enable the Baseten provider feature:

cargo add aisdk --features baseten

This installs AISDK with the Baseten provider enabled. Once you have enabled the Baseten provider, you can use all aisdk features with it.

Create a Provider Instance

To create a provider instance, call Baseten::model_name(), where model_name is the Baseten model you want to use. Model names are exposed as snake-case methods.

use aisdk::providers::Baseten;

let baseten = Baseten::qwen_qwen3_coder_480b_a35b_instruct();

This initializes the provider with:

Model: "Qwen/Qwen3-Coder-480B-A35B-Instruct"
API key from environment (if set with BASETEN_API_KEY)
Baseten's default base URL (https://inference.baseten.co/v1)

Basic Text Generation

Example using LanguageModelRequest for text generation.

use aisdk::{
    core::LanguageModelRequest,
    providers::Baseten,
};

#[tokio::main]
async fn main() -> Result<(), Box<dyn std::error::Error>> {

    let baseten = Baseten::qwen_qwen3_coder_480b_a35b_instruct();

    let response = LanguageModelRequest::builder()
        .model(baseten)
        .prompt("Write a short poem about Rust.")
        .build()
        .generate_text()
        .await?;

    println!("Response text: {:?}", response.text());
    Ok(())
}

Provider Settings

You can customize provider configuration using Baseten::builder()

API Key

let baseten = Baseten::<QwenQwen3Coder480bA35bInstruct>::builder()
    .api_key("your-api-key")
    .build()?;

If not specified, AISDK uses the BASETEN_API_KEY environment variable.

Base URL

Useful when routing through a proxy, gateway, or self-hosted compatible endpoint.

let baseten = Baseten::<QwenQwen3Coder480bA35bInstruct>::builder()
    .base_url("https://inference.baseten.co/v1")
    .build()?;

Path (Full URL Override)

Use .path(...) to override the full request URL instead of only the base URL.

let baseten = Baseten::<QwenQwen3Coder480bA35bInstruct>::builder()
    .path("https://full-url.example/v1/chat/completions")
    .build()?;

Provider Name

For logging, analytics, and observability.

let baseten = Baseten::<QwenQwen3Coder480bA35bInstruct>::builder()
    .provider_name("Baseten")
    .build()?;

Full Custom Configuration Example

let baseten = Baseten::<QwenQwen3Coder480bA35bInstruct>::builder()
    .api_key("your-api-key")
    .base_url("https://inference.baseten.co/v1")
    .path("https://full-url.example/v1/chat/completions")
    .provider_name("Baseten")
    .build()?;

Dynamic Model Selection

For runtime model selection (e.g., loading models from config files), use DynamicModel:

Using model_name() Method with Default Settings

use aisdk::providers::Baseten;

// Specify model as a string at runtime
let baseten = Baseten::model_name("Qwen/Qwen3-Coder-480B-A35B-Instruct");

Using Builder Pattern with Custom Settings

use aisdk::{
    core::DynamicModel,
    providers::Baseten,
};

let baseten = Baseten::<DynamicModel>::builder()
	.model_name("Qwen/Qwen3-Coder-480B-A35B-Instruct")
	.api_key("your-api-key")
	.base_url("https://inference.baseten.co/v1")
	.path("https://full-url.example/v1/chat/completions")
	.provider_name("Baseten")
	.build()?;

Warning: When using DynamicModel, model capabilities are not validated at compile time. This means there's no guarantee the model supports requested features (e.g., tool calls, structured output). For compile-time safety, use the typed methods like Baseten::qwen_qwen3_coder_480b_a35b_instruct().

Next Steps

Take a deeper look at text generation features Generating Text / Streaming Text
Explore Structured Output for reliable agent data.
Learn how to create Custom Tools.
Learn more about Agents.

Baseten

On this page