Official

anthropic / claude-3.5-sonnet

Anthropic's most intelligent language model to date, with a 200K token context window and image understanding (claude-3-5-sonnet-20241022)

Warm

Public
259.2K runs
Priced per token
Commercial use
License

Run with an API

Playground API Examples README

Input

Run this model in Node.js with one line of code:

npx create-replicate --model=anthropic/claude-3.5-sonnet

or set up a project from scratch

Install Replicate’s Node.js client library:

npm install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import and set up the client:

import Replicate from "replicate";

const replicate = new Replicate({
  auth: process.env.REPLICATE_API_TOKEN,
});

Run anthropic/claude-3.5-sonnet using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

const input = {
  prompt: "Give me a recipe for pancakes that could feed all of California.",
  max_tokens: 8192,
  system_prompt: "",
  max_image_resolution: 0.5
};

for await (const event of replicate.stream("anthropic/claude-3.5-sonnet", { input })) {
  process.stdout.write(event.toString());
};

To learn more, take a look at the guide on getting started with Node.js.

Install Replicate’s Python client library:

pip install replicate

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Import the client:

import replicate

Run anthropic/claude-3.5-sonnet using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

# The anthropic/claude-3.5-sonnet model can stream output as it's running.
for event in replicate.stream(
    "anthropic/claude-3.5-sonnet",
    input={
        "prompt": "Give me a recipe for pancakes that could feed all of California.",
        "max_tokens": 8192,
        "system_prompt": "",
        "max_image_resolution": 0.5
    },
):
    print(str(event), end="")

To learn more, take a look at the guide on getting started with Python.

Set the REPLICATE_API_TOKEN environment variable:

export REPLICATE_API_TOKEN=<paste-your-token-here>

Find your API token in your account settings.

Run anthropic/claude-3.5-sonnet using Replicate’s API. Check out the model's schema for an overview of inputs and outputs.

curl -s -X POST \
  -H "Authorization: Bearer $REPLICATE_API_TOKEN" \
  -H "Content-Type: application/json" \
  -H "Prefer: wait" \
  -d $'{
    "input": {
      "prompt": "Give me a recipe for pancakes that could feed all of California.",
      "max_tokens": 8192,
      "system_prompt": "",
      "max_image_resolution": 0.5
    }
  }' \
  https://api.replicate.com/v1/models/anthropic/claude-3.5-sonnet/predictions

To learn more, take a look at Replicate’s HTTP API reference docs.

Output

Here's a humorous yet enormous pancake recipe scaled up for California's population (approximately 40 million people): "California-Sized Pancake Recipe" Ingredients: * 20 million cups all-purpose flour (about 2,500 tons) * 40 million eggs (need about 3.3 million chickens working overtime) * 15 million cups milk (about 240,000 cows' daily production) * 10 million tablespoons baking powder * 5 million teaspoons salt * 5 million cups sugar * 5 million cups melted butter You'll need: * The entire Central Valley as a mixing bowl * Several helicopters with giant whisks * A tectonic plate-sized griddle * The Pacific Ocean converted to maple syrup Instructions: 1. Mix dry ingredients in Death Valley 2. Combine wet ingredients in Lake Tahoe 3. Pour batter using the California Aqueduct 4. Cook on the San Andreas Fault line for even heat distribution 5. Flip using coordinated effort from entire US Air Force Warning: May cause minor earthquakes when flipping. Not responsible for syrup-related flooding. Serves: 40 million Californians, plus leftovers for Nevada Note: This is obviously a silly recipe - it would be logistically impossible to make and cook a pancake this large! 😄

{
  "completed_at": "2025-02-12T16:19:30.270847Z",
  "created_at": "2025-02-12T16:19:24.273000Z",
  "data_removed": false,
  "error": null,
  "id": "8bwz4aszy5rm80cmzbttpakyfc",
  "input": {
    "prompt": "Give me a recipe for pancakes that could feed all of California.",
    "max_tokens": 8192,
    "system_prompt": "",
    "max_image_resolution": 0.5
  },
  "logs": "Running prediction...",
  "metrics": {
    "input_token_count": 21,
    "tokens_per_second": 54.00962598247224,
    "output_token_count": 323,
    "predict_time": 5.982320888,
    "total_time": 5.997847,
    "time_to_first_token": 0.017431468999999998
  },
  "output": [
    "Here's a hum",
    "orous yet enormous",
    " pancake recipe scale",
    "d up for California's population (",
    "approximately 40 million people",
    "):\n\n\"California-",
    "Sized Pancake",
    " Recipe\"\n\nIngredients",
    ":\n* 20",
    " million cups all-purpose flour (",
    "about 2,500",
    " tons)\n* 40",
    " million eggs (nee",
    "d about 3",
    ".3 million chickens working",
    " overtime)\n* 15",
    " million cups milk (about 240",
    ",000 cows'",
    " daily production)\n* 10",
    " million tablespoons b",
    "aking powder\n* 5 million",
    " teaspoons salt\n*",
    " 5 million cups sugar",
    "\n* 5 million cups mel",
    "ted butter\n\nYou'll nee",
    "d:\n* The entire Central",
    " Valley as a mixing bowl\n* Several",
    " helicopters with giant whisks\n* A",
    " tectonic plate-",
    "sized griddle\n*",
    " The Pacific Ocean converted to",
    " maple syrup\n\nInstructions",
    ":\n1. Mix dry",
    " ingredients in Death Valley",
    "\n2. Combine wet ingredients in Lake",
    " Tahoe\n3. Pour",
    " batter using the California",
    " Aqueduct\n4. Cook",
    " on the San Andreas Fault line",
    " for even heat distribution\n5.",
    " Flip using coordinated effort from entire",
    " US Air Force\n\nWarning",
    ": May cause minor earthquakes when",
    " flipping. Not responsible for sy",
    "rup-related flooding.\n\nServes",
    ": 40 million Californ",
    "ians, plus leftovers for Nevada",
    "\n\nNote: This is obviously",
    " a silly recipe - it woul",
    "d be logistically impossible to make an",
    "d cook a pancake this large!",
    " 😄"
  ],
  "started_at": "2025-02-12T16:19:24.288526Z",
  "status": "succeeded",
  "urls": {
    "stream": "https://stream-b.svc.ric1.c.replicate.net/v1/streams/fj6cmzuqif63sszp4whb5gdzmptgjvqz3r6wyge2gqo7wl4gxzja",
    "get": "https://api.replicate.com/v1/predictions/8bwz4aszy5rm80cmzbttpakyfc",
    "cancel": "https://api.replicate.com/v1/predictions/8bwz4aszy5rm80cmzbttpakyfc/cancel"
  },
  "version": "hidden"
}

Generated in

6.0 seconds

Input tokens

Output tokens

320

Tokens per second

54.01 tokens / second

Time to first token

17 milliseconds

Tweak it Report

Pricing

This language model is priced by how many input tokens are sent as inputs and how many output tokens are generated.

Type	Per unit	Per $1
Input	$3.75 / 1M tokens or 260K tokens / $1
Output	$18.75 / 1M tokens or 53K tokens / $1

For example, for $10 you can run around 762 predictions where the input is a sentence or two (15 tokens) and the output is a few paragraphs (700 tokens).

Check out our docs for more information about how per-token pricing works on Replicate.

Readme

Claude 3.5 Sonnet

Claude 3.5 Sonnet is Anthropic’s latest frontier large language model, offering state-of-the-art performance with significantly improved speed and cost-effectiveness compared to previous models.

This Replicate model is built on claude-3-5-sonnet-20241022, hosted on Anthropic and Vertex APIs.

Key features

Best-in-class performance on graduate-level reasoning (GPQA), undergraduate knowledge (MMLU), and coding (HumanEval)
2x faster inference speed compared to Claude 3 Opus
200K token context window
Advanced vision capabilities for interpreting images, charts, and graphics
Exceptional natural language understanding and generation
Industry-leading safety and privacy protections

Use cases

Complex customer support
Multi-step workflow orchestration
Code generation and debugging
Legacy code migration
Visual analysis and interpretation
Content generation with natural tone
Academic and research assistance

Limitations

Model does not learn from user interactions
No persistent memory between sessions
Subject to hallucinations and potential biases
Cannot access real-time information
Limited to text and image inputs

Safety

Rigorous safety testing and red teaming conducted
Evaluated by UK and US AI Safety Institutes
Enhanced safety mechanisms based on expert feedback
No training on user data without explicit consent
Built-in safeguards against misuse

Getting started

To use Claude 3.5 Sonnet via Replicate:

import replicate

output = replicate.run(
    "anthropic/claude-3.5-sonnet",
    input={
        "prompt": "Your prompt here"
    }
)

Privacy policy & license

Data from this model is sent to Anthropic and Google Cloud Vertex AI.

Usage of this model is subject to Anthropic’s terms of service. Please refer to their website for full terms and conditions.