Claude 3.5 Sonnet - In-Depth Overview

Anthropic · Claude

outdated

Model ID: claude-3-5-sonnet-20241022

Get all the details on Claude 3.5 Sonnet, an AI model from Anthropic. This page covers its token limits, pricing structure, key capabilities such as advanced_reasoning, multilingual, function_calling, available API code samples, and performance strengths.

Key Metrics

Input Limit

200K tokens

Output Limit

8.2K tokens

Input Cost

$3.00/1M

Output Cost

$15.00/1M

Sample API Code

import anthropic

client = anthropic.Anthropic(
    api_key="my_api_key",
)
message = client.messages.create(
    model="claude-3-5-sonnet-20241022",
    max_tokens=1024,
    messages=[
        {"role": "user", "content": "Hello, Claude"}
    ]
)
print(message.content)

Required Libraries

anthropic

@anthropic-ai/sdk

Benchmarks

Benchmark	Score	Source	Notes
lmarena text	1280	Open LLM Leaderboard	Rank 38 overall on May 20, 2025.
lmarena webdev	1238	Open LLM Leaderboard	Rank 6 overall on May 20, 2025.
lmarena vision	1183	Open LLM Leaderboard	Rank 17 overall on May 20, 2025.
livebench global average	57.94	LiveBench	Scores from LiveBench-2025-04-25.
livebench reasoning average	43.22	LiveBench	Scores from LiveBench-2025-04-25.
livebench coding average	73.90	LiveBench	Scores from LiveBench-2025-04-25.
livebench mathematics average	50.54	LiveBench	Scores from LiveBench-2025-04-25.
livebench data analysis average	56.19	LiveBench	Scores from LiveBench-2025-04-25.
livebench language average	54.48	LiveBench	Scores from LiveBench-2025-04-25.
livebench if average	69.30	LiveBench	Scores from LiveBench-2025-04-25.
aime 2024	16 %	Vellum LLM Leaderboard	Updated 17 April 2025.
gpqa	65 %	Vellum LLM Leaderboard	Updated 17 April 2025.
swe bench	49 %	Vellum LLM Leaderboard	Updated 17 April 2025.
math 500	78 %	Vellum LLM Leaderboard	Updated 17 April 2025.
bfcl	56.46 %	Vellum LLM Leaderboard	Updated 17 April 2025.
alder polyglot	51.6 %	Vellum LLM Leaderboard	Updated 17 April 2025.

Notes

Described as 'Our previous intelligent model'. It has a 'latest' alias (claude-3-5-sonnet-latest) which points to this version. Does not support extended thinking. Supports multilingual input/output, vision, prompt caching, streaming messages, batch processing, and citations. Tool use is supported.