Claude 3 Haiku, Anthropic's fastest model, is now officially available
Anthropic officially launched Claude 3 Haiku on the claude.ai platform, the Claude API, and Amazon Bedrock, with upcoming Google Cloud Vertex AI availability. The vision-enabled model is fast, affordable, and is particularly adept at math reasoning and coding.
Anthropic recently launched the Claude 3 family of models, beginning with the general availability of Sonnet and Opus at claude.ai for free and Pro tiers of the web-based experience, respectively. The company also announced that the models would be accessible via its generally available Claude API. Most importantly, Anthropic announced that Haiku, its fastest and most affordable vision-enabled model, would join Sonnet and Opus on both platforms shortly. Exactly a week later, Anthropic has finally announced the launch of Haiku for Claude Pro, Claude API, and Amazon Bedrock users, with the model available starting March 13. Additionally, Haiku will be available on the Google Cloud Vertex AI soon.
Claude 3 Haiku is a fast, vision-enabled model that delivers highly competitive performance across the reported benchmarks tested. Like its siblings, the model is particularly adept at math reasoning and coding, delivering superior scores than GPT-3.5 and Gemini 1.0 Pro.
Haiku can process 21K tokens (about 30 pages of text) per second for prompts under 32K tokens. Haiku's 1:5 input-to-output token ratio pricing model ($0.25/$1.25) was set with the understanding that enterprises will use the model to analyze substantial amounts of data, especially documents and images. To showcase Haiku's level of cost-effectiveness, Anthropic stated that the model can process 400 Supreme Court cases (10K tokens each) or 2,500 images (1.6 tokens each) for just one US dollar.
Since Haiku is a model geared towards business and enterprise use cases, it has been subject to several layers of security and robustness testing. Haiku has also been set up with additional layers of defense, such as continuous systems monitoring, endpoint hardening, secure coding practices, and strong data encryption protocols. Moreover, Anthropic states it performs continuous audits and penetration testing rounds on its models to identify and prevent vulnerabilities. The full technical details for Haiku and the other Claude 3 models are available in the Claude 3 model card.
