![Anthropic Releases New GenAI Models with Claude 3](http://media.assettype.com/cdomagazine%2F2024-03%2Fc58558f1-d20d-419e-a953-f2ce1ef6b1ca%2Fsource_anthropic.png?w=480&auto=format%2Ccompress&fit=max)
Amazon and Google-backed AI startup Anthropic has unveiled its new lineup of Claude 3 models, claiming superiority over Google's Gemini and OpenAI's ChatGPT across various benchmarks.
The family includes three generative AI variants: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus. These models boast enhanced multimodality, increased accuracy, improved contextual understanding, and faster responses, with a distinct focus on addressing challenging queries.
All three models boast a 200k context window that can handle as high as one million tokens.
Sonnet and Opus are currently available for purchase, with a free version of Claude accessible on Anthropic's website. The release date for Haiku is pending.
Sonnet is also currently accessible on Amazon Bedrock, and it is in the private preview phase on Google Cloud's Vertex AI Model Garden. Additionally, Opus and Haiku are set to be made available on both platforms shortly.
Anthropic addresses the limitations of older Claude versions, highlighting the Claude 3 family's nuanced response approach, especially when faced with complex or boundary-pushing questions. The trio promises an overall performance boost, with Opus positioned as the pinnacle, exhibiting nearly human-like comprehension for intricate tasks.
Opus, considered the top-tier model, undergoes a rigorous "Needle In a Haystack" (NIAH) evaluation to test data recall ability, displaying commendable performance. Opus displays a near-perfect 99% recall ability when processing long context prompts.
Anthropic claims Opus surpasses GPT-4 in areas such as solving maths problems, generating code, and displaying superior reasoning skills. However, Opus does exhibit some quirks, including occasional inaccuracies and a slightly delayed response time compared to Claude 2.
While Opus shines in large-scale operations, Haiku excels in quick replies and information retrieval from unstructured data. Sonnet, a larger-scale model, aids in time-saving for menial tasks and parsing text from images. Each model caters to specific use cases, demonstrating versatility within the Claude 3 family.