Major Release Anthropic LLM Proprietary

Anthropic Claude 3 Haiku

Fastest and cheapest Claude 3 — sub-second latency at $0.25/M

Released 2024-03-04 · ~25B · 200K tokens context · knowledge cutoff 2023-08

Overview

Fastest and most compact model in the Claude 3 family

Specifications

DeveloperAnthropic
Release date2024-03-04
Model typeLLM
Parameters~25B
ArchitectureTransformer
Context window200K tokens
Max output4096 tokens
Knowledge cutoff2023-08
LicenseProprietary
Input modalitiestext, image
Output modalitiestext

Benchmarks

Benchmark Claude 3 Haiku
MMLU 75.2%
HumanEval 75.9%
MATH 38.9%

Availability

API Only Anthropic API AWS Bedrock Google Vertex AI

Official references