Anthropic Unveils Claude 3 AI Models, Claims It Beats GPT-4 and Gemini

Anthropic has introduced its new family of artificial intelligence (AI) models called Claude 3. The third generation of the company’s AI-powered chatbot now comes in three separate versions — Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus — where Opus is the most capable model, followed by Sonnet and Haiku. The company has also shared results from benchmark testing of the chatbot and has claimed that the AI bot outperforms both OpenAI’s GPT-4 and Google’s Gemini 1.0 Ultra models.

Announcing the three AI models in a post, Anthropic said, “All Claude 3 models show increased capabilities in analysis and forecasting, nuanced content creation, code generation, and conversing in non-English languages like Spanish, Japanese, and French.” Interestingly, this is the first multimodal model created by the AI firm, meaning the chatbot can receive input in both text and image formats. However, Claude 3 cannot generate images.

Anthropic’s Claude 3 AI models: Details

The multimodal status of the AI model is not its only highlight. Claude 3’s context window has also seen an upgrade, now extending to up to 2,00,000 tokens for Haiku and Sonnet. In some specific cases, Opus can also reach up to 1 million tokens for some specific tasks. Interestingly, the standard context length of 2 lakhs is more than the recently announced Gemini 1.5 Pro’s 128,000 tokens (although, it also has a special model that can reach up to 1 million tokens). For the unversed, the context window is a frame through which the AI views its data to process information. The larger this window, the more surrounding information it can see to build a better context and understanding of the queried topic.

The company also shared Claude 3’s benchmarking results. In terms of general capability, the company claimed that the Opus model outperformed both GPT-4 and Gemini 1.0 Ultra in popular tests such as MMLU, HumanEval, GSM8K, and HellaSwag. Anthropic also claimed that the AI models can share near-instant results for even complex research papers.

Coming to the individual models within Claude 3, Haiku is the smallest AI model which can also produce the fastest responses. Sonnet falls in the middle and can offer more intelligent responses than Haiku at still comparatively lower costs. Opus is the most intelligent model. “It can navigate open-ended prompts and sight-unseen scenarios with remarkable fluency and human-like understanding,” the company added.

Both Opus and Sonnet are now available to developers as API and users via its chatbot platform. Sonnet will be powering the free version on Claude.ai, whereas Opus will be available with the Claude Pro subscription, which costs $20 (roughly Rs. 1,650) a month. Additionally, Sonnet is also available through Amazon Bedrock and Google’s Vertex AI in private preview. The company has stated that Opus and Sonnet will be added soon as well.