Anthropic claims: its Claude 3 model surpasses GPT-4, with “almost human” capabilities

Available in three versions, the large Claude 3 language model is now accessible and further pushes the limits of generative artificial intelligence.

Anthropic is both a competitor and an ally of the digital giants in the large language models (LLM) sector. The company has just launched a new family of LLMs, whose performance is through the roof, surpassing OpenAI’s GPT-4.

Three new generative AI models

Three models are announced: Claude 3 Haiku, Claude 3 Sonnet and Claude 3 Opus. The Opus model is the most advanced in the series, “presenting quasi-human levels of understanding and ease on complex tasks, approaching general intelligence”praises Anthropic.

Knowledge, ability to solve a mathematical problem, code generation, ability to simulate reasoning… Claude 3 Opus obtains better scores than GPT-4 on many key criteria on the main benchmark tools.

The company also promises near-instant results for tasks like live customer chat, auto-completion, and data extraction.

Faster and more relevant

Claude 3 models can process photos, tables, graphs and technical diagrams. They also have a better understanding of the nuances to generate results more in line with user expectations, and are much less likely to refuse to respond to even innocuous prompts.

Anthropic says it significantly improved Claude’s responses to difficult open-ended questions and reduced the levels of incorrect answers. For increased reliability, Claude 3 will soon benefit from the possibility of citing a document by pointing to specific sentences.

All three new models also manage to follow complex, multi-step instructions. Claude 3 Sonnet and Claude 3 Opus are already available from the Claude AI chatbot and Anthropic APIs. Claude 3 Haiku will follow soon.

Source : Anthropic