ChatAnthropic

Anthropic Claude LLM provider for Browser Use

Overview

ChatAnthropic provides integration with Anthropic’s Claude models, including Claude 3.5 Sonnet, Claude 3 Opus, and other Claude family models.

Basic Usage

from browser_use import Agent, ChatAnthropic
import asyncio

async def main():
    llm = ChatAnthropic(model='claude-sonnet-4-0', temperature=0.0)
    agent = Agent(
        task="Find the number 1 post on Show HN",
        llm=llm,
    )
    await agent.run()

if __name__ == "__main__":
    asyncio.run(main())

Configuration

Required Parameters

model · str · required

Claude model to use. Common options:

• claude-sonnet-4-0: Latest Claude 3.5 Sonnet
• claude-3-5-sonnet-20241022: Specific Sonnet version
• claude-3-opus-20240229: Most powerful Claude 3 model
• claude-3-sonnet-20240229: Balanced performance
• claude-3-haiku-20240307: Fast and cost-effective

Model Parameters

max_tokens · int · default: "8192"

Maximum tokens to generate in the response.

temperature · float · default: "None"

Sampling temperature (0.0 to 1.0). Controls randomness in responses.

top_p · float · default: "None"

Nucleus sampling parameter (0.0 to 1.0).

seed · int · default: "None"

Random seed for deterministic output (experimental).

Client Parameters

api_key · str · default: "None"

Anthropic API key. Defaults to ANTHROPIC_API_KEY environment variable.

Get your API key at console.anthropic.com

auth_token · str · default: "None"

Alternative authentication token.

base_url · str · default: "None"

Custom base URL for Anthropic API.

timeout · float · default: "None"

Request timeout in seconds or httpx.Timeout object.

max_retries · int · default: "10"

Maximum number of retries for failed requests.

default_headers · dict · default: "None"

Additional headers to include in all requests.

default_query · dict · default: "None"

Additional query parameters for all requests.

http_client · httpx.AsyncClient · default: "None"

Custom async HTTP client.

Advanced Usage

Structured Output with Tool Calling

ChatAnthropic uses Claude’s tool calling feature for structured outputs:

from browser_use import Agent, ChatAnthropic
from pydantic import BaseModel

class Article(BaseModel):
    title: str
    author: str
    date: str
    summary: str

llm = ChatAnthropic(
    model='claude-sonnet-4-0',
    temperature=0.0,
)

agent = Agent(
    task="Extract article information",
    llm=llm,
    output_model_schema=Article,
)

result = await agent.run()
print(result.structured_output)  # Article instance

Prompt Caching

Claude supports prompt caching for frequently used context:

from browser_use import Agent, ChatAnthropic

llm = ChatAnthropic(
    model='claude-sonnet-4-0',
    temperature=0.0,
)

# The agent automatically uses cache_control for system prompts
agent = Agent(
    task="Complex task with large context",
    llm=llm,
)

Prompt caching can significantly reduce costs for repeated contexts. Cache hits are reflected in usage.prompt_cached_tokens.

Using with Custom System Prompts

from browser_use import Agent, ChatAnthropic

llm = ChatAnthropic(
    model='claude-sonnet-4-0',
    temperature=0.0,
)

agent = Agent(
    task="Your task",
    llm=llm,
    extend_system_message="Additional instructions for Claude.",
)

Environment Setup

.env

ANTHROPIC_API_KEY=your_api_key_here

Error Handling

from browser_use import Agent, ChatAnthropic
from browser_use.llm.exceptions import (
    ModelProviderError,
    ModelRateLimitError
)

try:
    llm = ChatAnthropic(model='claude-sonnet-4-0')
    agent = Agent(task="Your task", llm=llm)
    result = await agent.run()
except ModelRateLimitError as e:
    print(f"Rate limit exceeded: {e.message}")
except ModelProviderError as e:
    print(f"API error: {e.message}")
    print(f"Status code: {e.status_code}")

Properties

provider

Returns the provider name: "anthropic"

llm = ChatAnthropic(model='claude-sonnet-4-0')
print(llm.provider)  # "anthropic"

name

Returns the model name.

llm = ChatAnthropic(model='claude-sonnet-4-0')
print(llm.name)  # "claude-sonnet-4-0"

Methods

get_client()

Returns an AsyncAnthropic client instance.

llm = ChatAnthropic(model='claude-sonnet-4-0')
client = llm.get_client()
# Use client directly for advanced operations

ainvoke()

Asynchronously invoke the model with messages.

from browser_use.llm.messages import SystemMessage, UserMessage

llm = ChatAnthropic(model='claude-sonnet-4-0')

messages = [
    SystemMessage(content="You are a helpful assistant"),
    UserMessage(content="What is Browser Use?")
]

response = await llm.ainvoke(messages)
print(response.completion)     # String response
print(response.usage)          # Token usage
print(response.stop_reason)    # Why generation stopped

Parameters

• messages (list[BaseMessage]): List of messages
• output_format (type[T] | None): Optional Pydantic model for structured output

Returns

ChatInvokeCompletion[T] | ChatInvokeCompletion[str] with:

• completion: Response content (string or structured output)
• usage: Token usage including:
  • prompt_tokens: Total input tokens (including cached)
  • completion_tokens: Output tokens
  • prompt_cached_tokens: Tokens retrieved from cache
  • prompt_cache_creation_tokens: Tokens written to cache
• stop_reason: Completion reason (end_turn, max_tokens, stop_sequence, tool_use)

Token Usage

Claude’s token counting is unique:

response = await llm.ainvoke(messages)

if response.usage:
    print(f"Prompt tokens: {response.usage.prompt_tokens}")
    print(f"Cached tokens: {response.usage.prompt_cached_tokens}")
    print(f"Cache writes: {response.usage.prompt_cache_creation_tokens}")
    print(f"Completion: {response.usage.completion_tokens}")

Anthropic includes cached tokens in the total prompt token count. Check prompt_cached_tokens for actual cache hits.

Model Capabilities

Claude 3.5 Sonnet

• Best balance of intelligence and speed
• 200K token context window
• Strong coding and analysis capabilities
• Native vision support

Claude 3 Opus

• Most powerful Claude model
• Best for complex tasks
• 200K token context window
• Highest accuracy

Claude 3 Haiku

• Fastest and most cost-effective
• Great for simple tasks
• 200K token context window
• Instant responses

Related

• ChatBrowserUse - Recommended provider
• ChatOpenAI
• ChatGoogle