Claude Logo

Claude 3 Haiku

Fastest and most compact Claude 3 model; optimized for near-instant responses; excels at high-volume cost-sensitive tasks.

Claude 3 Haiku released March 4 2024 as part of the Claude 3 family served as the smallest and fastest model in the lineup. Designed for near-instant responsiveness it achieved 123.1 tokens per second output speed with 0.71 second latency. Ideal for live customer interactions real-time translations content moderation inventory management and knowledge extraction from unstructured data. The model introduced vision capabilities allowing it to process images and visual content. Priced significantly below Sonnet and Opus it enabled seamless AI experiences mimicking human interaction speed while maintaining reasonable accuracy for simpler tasks. Haiku 3 established the tier system that became central to Anthropic's model strategy allowing users to choose between speed cost and capability based on their specific needs.

Reviews

No Reviews Yet

Be the first to share your experience with this AI tool

More models from Claude

Most intelligent Claude model; 80.9% SWE-bench; effort parameter for compute control; Infinite Chats feature.

Balanced mid-tier model with 2x speed improvement over Claude 2; strong enterprise performance at moderate cost.

Flagship Claude 3 model; 200K context (expandable to 1M); outperformed GPT-4 on most benchmarks at launch.

Outperformed Claude 3 Opus at half the cost; introduced Artifacts feature; benchmark leader in coding and reasoning.

Major October 2024 upgrade; introduced computer use capability; enhanced coding and agentic task performance.

Fast efficient model matching Claude 3 Opus performance; cost-effective option for scaled deployments.

First hybrid reasoning model; toggleable extended thinking mode; 70.3% SWE-bench score; 128K output tokens.

Claude 4 family workhorse; superior coding and reasoning over 3.7; hybrid reasoning with tool use support.

Most powerful Claude 4 model; ASL-3 safety classification; 72.5% SWE-bench; excels at sustained 7-hour coding sessions.

Incremental Opus 4 upgrade; 74.5% SWE-bench score; enhanced agentic reasoning and multi-file refactoring.

Best coding model in the world; 77.2% SWE-bench; 61.4% OSWorld for computer use; 30+ hour sustained focus.

Near-frontier performance at lowest cost; matches Sonnet 4 on coding; first Haiku with extended thinking.