Anthropic, an artificial intelligence (AI) company with support from industry giants like Amazon, Google, and Salesforce, has unveiled its latest and most powerful large language model (LLM) in the intensifying race for AI dominance. Founded by former OpenAI members, Anthropic released the Claude 3 model family on Monday, featuring three progressively advanced models: Claude 3 Haiku, Claude 3 Sonnet, and Claude 3 Opus.
Anthropic asserted that Opus, the most sophisticated among the models, surpassed OpenAI’s GPT-4 and Google’s Gemini Ultra in various tests, including reasoning, basic mathematics, and knowledge at undergraduate and graduate levels. Opus and Sonnet are now accessible on claude.ai and the Claude API, available globally in 159 countries, while the more compact Haiku and Sonnet models are expected to be released soon.
Opus stands out with the ability to summarize around 200,000 tokens, a significant leap compared to ChatGPT’s 3,000 tokens. Anthropic’s new models also introduce the capability to process images and documents, a novel feature in their lineup.
Formed in 2021 by siblings Dario and Daniela Amodei after departing OpenAI, Anthropic prioritizes safety concerns often overlooked by other entities. The company boasts dedicated teams addressing risks like misinformation, biological misuse, and election interference. Anthropic emphasizes its commitment to reducing biases in its models, with Claude 3 showing fewer biases than its predecessors.
As a public-benefit corporation, Anthropic is obligated to report how it contributes to the public’s welfare. Collaborations with companies like Airtable and Asana for A/B testing have played a role in refining the models, with reported improvements in latency and output quality. Eric Pelz, Head of Technology, AI at Asana, expressed enthusiasm, citing a 42% improvement in time-to-first-token and over a 200% improvement in tokens per second.
