Claude 3: Anthropic creates new AI models that are superior to those of Google and OpenAI

March 5, 2024  12:13

Anthropic, a company founded by former OpenAI employees, has announced the launch of a new generation of artificial intelligence (AI) models called Claude 3. These models are the result of meticulous work in machine learning and represent significant progress compared to their predecessors and current counterparts from other developers, including OpenAI and Google.

Claude 3 stands out for its multimodality—the ability to understand and process both textual and visual data. This expansion of the AI model's functionality not only makes it more versatile but also opens up new possibilities for its application in various fields, from education to medicine.

The Claude 3 family includes three models: Haiku, Sonnet, and Opus. Opus, being the most advanced model, is described by the company as the most intelligent. Currently, Opus and Sonnet models are available to users through the official website and the application programming interface (API). The Haiku model, known for its high data processing speed with minimal costs, will be introduced to the public in the near future.

claude 3 chart .jpg (41 KB)

It's worth noting that due to insufficient contextual understanding, previous versions of Claude sometimes refused to respond to certain user queries, even if they were harmless. The new Claude 3 models possess improved contextual understanding and are less prone to refusal in responses, making them more adaptive and flexible in their operations.

Claude 3 demonstrates an outstanding ability to process and analyze complex materials, such as scientific articles with texts, diagrams, and graphs, in less than 3 seconds. This positions it among the fastest and most economical models on the market.

claude 3 chart 2 .jpg (260 KB)

In several benchmarks, the Opus model showed better results than OpenAI's counterparts, especially in tasks requiring master's level reasoning, where it outperformed GPT-4, showing a result of 50.4% compared to GPT-4's 35.7%. This confirms its high analytical capabilities and depth of understanding of the material.

Anthropic conducted the training of Claude 3 models using a mixed dataset, including not only publicly available sources but also internal data, achieving high-quality information processing. The training of models took place on the computing resources of Amazon AWS and Google Cloud, emphasizing the scale and significance of the project. Both Amazon and Google have invested significant funds in the development of Anthropic, indicating a high assessment of the potential of its new AI models.

  • Archive