On Monday, local time, artificial intelligence (AI) startup Anthropic launched Claude 3 big models that are better able to execute complex instructions.
The company claims that Claude 3 exhibits a near-human understanding of complex tasks and is one of the most powerful large models currently available.
The Claude 3 is further divided into three versions, called the Claude 3 Opus, Sonnet and Haiku, with the names suggesting the capabilities of each model, with the Opus being the most powerful and the Haiku the fastest and most cost-effective model on the market.
According to Anthropic, the most powerful Opus outperforms OpenAI’s GPT-4 and Google’s Gemini Ultra in industry benchmarks, such as knowledge level, deductive reasoning, and basic math.
Claude 3 is also a multimodal large model with powerful “visual capabilities,” so users can upload photos, charts, documents, and other types of data, analyze them, and ask questions.
Separately, Opus and Sonnet were made available to 159 countries and territories on Monday, while Haiku will launch in the coming weeks.
According to reports, Claude 3 can summarize about 150,000 words, while the previous version can only summarize 75,000 words. By comparison, OpenAI’s GPT-4 can process about 25,000 words of text.
Anthropic founders Daniela Amodei and Dario Amodei are both former OpenAI executives, the latter being OpenAI’s vice President of research.
They founded the company in 2021 with the goal of developing generative AI models to compete with OpenAI. Its Anthropic is currently the hottest star startup in addition to OpenAI.
Over the past year, Anthropic has closed five different financing deals for a total of about $7.3 billion, valuing the company at $18.4 billion at the end of December.
Its investors include Google, Salesforce, and Amazon, and its products compete directly with ChatGPT in both the enterprise and consumer worlds.
According to Daniela Amodei, Claude 3 has a more detailed understanding of the problem, because of security considerations, Claude 2 often chooses to refuse to answer some sensitive questions, while Claude 3 has a better understanding of the risk and can answer more questions.
“At least at this point in time, Claude 3 is the Rolls-Royce of the big models,” Dario Amodei said in an interview.
However, Dario Amodei also pointed out that although Claude 3 is a multimodal large model, it will only examine and analyze the image, and will not provide the ability to directly generate the image, because the market demand has been greatly reduced, and customers have not asked for such a function.