Anthropic, a leading artificial intelligence startup, today announced the Claude 3 series of AI models. It is designed to meet the diverse needs of enterprise customers with a balance of intelligence, speed, and cost efficiency. The lineup includes his three models: Opus, Sonnet, and his upcoming Haiku. From the report: The centerpiece of the lineup is Opus, which Anthropic claims is more powerful than any other AI system on the market and outperforms rivals OpenAI and Google's leading models. Masu. “Opus can handle a wide range of tasks and performs them extremely well,” Anthropic co-founder and CEO Dario Amodei said in an interview with VentureBeat. Amodei explained that Opus outperforms top AI models such as GPT-4, GPT-3.5, and Gemini Ultra in a wide range of benchmarks. This includes topping the leaderboards in academic benchmarks such as his GSM-8k for mathematical reasoning and his MMLU for expert-level knowledge.
“It seems like they're outperforming everyone and getting scores on some tasks that we've never seen before,” Amodei said. Although companies such as Anthropic and Google do not publish the full parameters of their leading models, benchmark results reported by both companies show that Opus is comparable in core functionality to leading alternatives such as GPT-4 and Gemini. It suggests that it does or exceeds that. This sets a new high-water mark for commercially available conversational AI, at least in theory. Designed for complex tasks that require advanced reasoning, Opus stands out in the Anthropic lineup for its superior performance. A mid-range model, Sonnet provides businesses with a more cost-effective solution for routine data analysis and knowledge tasks, maintaining high performance without the hefty price tag of flagship models. . Haiku, on the other hand, is designed to be fast and economical, making it suitable for applications such as consumer chatbots where responsiveness and cost are important factors. Amodei told VentureBeat that he expects Haiku to be generally available within “weeks, not months.”