Model

Claude 3 Sonnet

CLOSED

Anthropic · Claude 3 family · released Mar 4, 2024

Balanced mid-tier of the Claude 3 family; roughly 2x faster than Claude 2 at a fraction of the cost.

ReasoningCodingVisionFunction callingTool useAgentic

821.3

Elo · rank #251

Parameters

Undisclosed

Active params

—

Context

200K tokens

Architecture

Dense multimodal Transformer (parameter count undisclosed)

License

Proprietary (Anthropic Commercial Terms of Service)

Languages

—

API price (in/out)

$3 / $15

Modalities

text · vision

Benchmark results

Bar shows position within the tracked field; marker = field best

AI2DVision88.7%#23

best: Molmo 72B · 96.3%

AIMEMath1.8%#151

best: GPT-5.2 · 100.0%

AMCMath18.0%#8

best: Falcon-H1 34B · 69.4%

ARC-ChallengeReasoning93.2%#15

best: Llama 3.1 405B · 96.9%

BIG-Bench HardReasoning82.9%#24

best: ERNIE 4.5 300B-A47B · 94.3%

ChartQAVision81.1%#32

best: MiniMax-VL-01 · 91.7%

DocVQAVision89.5%#34

best: Qwen2-VL-72B · 96.5%

DROPReasoning78.9#30

best: Hunyuan-T1 · 93.1

GPQA DiamondReasoning40.4%#216

best: GPT-5.6 · 94.6%

GSM8KMath92.3%#28

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning89.0%#10

best: Claude 3 Opus · 95.4%

HumanEvalCoding73.0%#74

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning81.1%#89

best: Gemma 4 26B A4B · 98.5%

LogicKorHuman preference8.36#13

best: GPT-4o · 9.33

MATH-500Math43.1%#144

best: GPT-5 · 99.4%

MathVistaVision47.9%#55

best: Seed 2.1 Pro · 90.7%

MBPPCoding79.4%#16

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MGSMMath83.5%#24

best: OpenAI o4-mini · 93.7%

MMLU-ProKnowledge54.9%#124

best: Claude Fable 5 · 91.5%

MMLUKnowledge79.0%#98

best: OpenAI o3 · 92.9%

MMMLUKnowledge69.0%#43

best: Gemini 3.1 Pro · 92.6%

MMMUVision53.1%#89

best: Claude Fable 5 · 89.3%

PubMedQAKnowledge78.3%#2

best: Palmyra Med 70B · 79.6%

RACE-HReasoning88.8%#3

best: Claude 3 Opus · 92.9%

SimpleQAKnowledge9.2%#40

best: GPT-4.5 · 62.5%

WinoGrandeReasoning75.1%#65

best: PaLM 2 · 90.9%

Run it locally

Closed weights — available via API only. No local deployment.

Input / M tok

Output / M tok

$15

Claude 3 family

Elo progression across releases

Claude 3 OpusMar 20241081.0 Claude 3 SonnetMar 2024821.3 Claude 3 HaikuMar 2024707.0

Resources

DOCSdocs.claude.com

API price $3/$15 · each benchmark row carries its own source badge (see methodology)