Anthropic releases its ‘smartest’ AI model

OpenAI rival Anthropic on Monday has released what it calls its smartest artificial intelligence model to date, especially when it comes to computer coding.

Along with Claude 3.7 Sonnet, the San Francisco-based company is making available in a limited research preview a digital “agent” called Claude Code tailored to be a tool for software developers.

Amazon-backed Anthropic described Claude Code as able to search and read code, edit files, run tests and more.

The release comes as AI companies are pushing out new products at a fast pace and with innovations quickly reproduced by rivals, often at a lower cost, raising concerns about finding a return on the massive investments.

Anthropic’s new model is “much stronger at coding, and particularly at taking over and doing really complicated coding tests,” Anthropic co-founder and chief science officer Jared Kaplan told AFP.

Aside from overall improved intelligence, the latest iteration of Claude has a “hybrid” reasoning model that lets users get quick answers to questions or have it spend time mulling complex queries and share steps in the process, according to Kaplan.

The improvement enables Claude to better follow instructions and handle more sophisticated analyses, he added.

Since OpenAI released ChatGPT in late 2022, the race has been on to lead in a technology predicted to change the way people live and work.

The new AI model makes its debut as Anthropic works on finalizing a $3.5 billion ($6.1b) funding round that values the start-up at $61.5b in a huge leap from its prior valuation, according to a Wall Street Journal report.

Anthropic declined to comment on the report.

AI models have moved beyond generating images, videos or written works to providing “agents” specialising in fields or tasks.

OpenAI released a version of ChatGPT about six months ago that shared its “thinking” process, but Anthropic followed that by enabling its Claude model to command computers as people do.

OpenAI responded with the recent release of its first AI agent called Operator with similar capabilities.

Anthropic, which was founded in 2021 by former OpenAI employees, and its arch-rival are striving to stand out in an increasingly crowded market.

“We try very hard to make model improvements grounded in customer problems,” said Anthropic chief product officer Mike Krieger.

“When it’s just newer, better, faster, it’s not as impactful. We try to hear what people are saying and have the next model serve those needs.”

Amazon has invested a total of $8b in Anthropic, while Google-parent Alphabet has invested $2b in the startup.

- Agence France-Presse