Model

Claude 3 Haiku

CLOSED

Anthropic · Claude 3 family · released Mar 13, 2024

Fastest, cheapest Claude 3; vision-capable with near-instant responses for its intelligence class.

ReasoningCodingVisionFunction callingTool useAgentic

707.0

Elo · rank #264

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)

$0.25 / $1.25

Modalities

text · vision

Benchmark results

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

AI2DVision86.7%#27

best: Molmo 72B · 96.3%

AIMEMath0.8%#152

best: GPT-5.2 · 100.0%

AMCMath32.0%#7

best: Falcon-H1 34B · 69.4%

ARC-ChallengeReasoning89.2%#28

best: Llama 3.1 405B · 96.9%

BIG-Bench HardReasoning73.7%#45

best: ERNIE 4.5 300B-A47B · 94.3%

ChartQAVision81.7%#29

best: MiniMax-VL-01 · 91.7%

DocVQAVision88.8%#36

best: Qwen2-VL-72B · 96.5%

DROPReasoning78.4#31

best: Hunyuan-T1 · 93.1

GPQA DiamondReasoning33.3%#249

best: GPT-5.6 · 94.6%

GSM8KMath88.9%#47

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning85.9%#32

best: Claude 3 Opus · 95.4%

HumanEvalCoding75.9%#63

best: Claude Opus 4.5 · 99.4%

IFEvalReasoning77.2%#100

best: Gemma 4 26B A4B · 98.5%

LogicKorHuman preference8.26#14

best: GPT-4o · 9.33

MATH-500Math38.9%#149

best: GPT-5 · 99.4%

MathVistaVision46.4%#56

best: Seed 2.1 Pro · 90.7%

MBPPCoding80.4%#13

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

MGSMMath75.1%#30

best: OpenAI o4-mini · 93.7%

MMLU-ProKnowledge49.0%#133

best: Claude Fable 5 · 91.5%

MMLUKnowledge75.2%#123

best: OpenAI o3 · 92.9%

MMMLUKnowledge65.2%#47

best: Gemini 3.1 Pro · 92.6%

MMMUVision50.2%#96

best: Claude Fable 5 · 89.3%

PubMedQAKnowledge76.0%#4

best: Palmyra Med 70B · 79.6%

RACE-HReasoning87.0%#6

best: Claude 3 Opus · 92.9%

SimpleQAKnowledge8.2%#41

best: GPT-4.5 · 62.5%

WinoGrandeReasoning74.2%#73

best: PaLM 2 · 90.9%

Run it locally

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

Input / M tok

$0.25

Output / M tok

$1.25

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 $0.25/$1.25 · each benchmark row carries its own source badge (see methodology)