Anthropic, an artificial intelligence company founded by several former OpenAI employees, says its new Claude 3 family of artificial intelligence models performs as well or better than competitors from Google and OpenAI. Unlike previous versions, Claude 3 is a multimodal model capable of understanding text and image input.
According to the company, Claude 3 will answer more questions, understand longer instructions, and be more accurate. Claude 3 can understand more context, meaning it can process more information. The family includes Claude 3 Haiku, Claude 3 Sonnet and Claude 3 Opus, where Opus is the largest and «the smartest» model.
Anthropic says that Opus and Sonnet are now available on claude.ai and as APIs. Haiku will be released soon, the junior version of Claude 3, according to the developers, is the «fastest and most economical model on the market», capable of reading a dense research article with charts and graphs in less than three seconds.
Previous versions of Claude refused to respond to some innocuous prompts, which the company believes «indicates a lack of understanding of the»’s context. The new models are free of this flaw.
Opus, the most advanced version, outperforms its competitors in all tests conducted by the company (see the results in the comparison table):
Anthropic trained Claude 3 models based on closed internal and third-party databases as well as publicly available data. The new models were trained using Amazon AWS and Google Cloud hardware. Both companies have invested in Anthropic, with Amazon investing $4 billion in the company. Claude 3 will be available in the AWS Bedrock model library and in Google’s Vertex AI.