Jaque Silva | Nurphoto | Getty Images
As heavily funded startups and tech giants hustle to get any lead they can in artificial intelligence, Anthropic says it’s developed the company’s “most intelligent” AI model yet.
The Amazon-backed startup on Monday unveiled Claude 3.7 Sonnet. What makes it unique is its so-called hybrid model, which combines an ability to reason — or stopping to think about complex answers — with a traditional model that spits out answers in real time.
“This model has all the capabilities wrapped together — we want one coherent AI that can help with everything,” Anthropic co-founder and science chief Jared Kaplan told CNBC in an interview. “There’s an advantage in simplicity for our customers.”
Anthropic says it’s the only “hybrid” model of its kind available on the market, and will go live immediately. Kaplan likened it to the way the human brain operates. Some questions require deep thinking, some require quick responses. But Anthropic is looking to integrate both capabilities, rather than have an entirely separate model for both.
The move could give Anthropic a much-needed edge against rival OpenAI, and megacap tech companies that are all investing heavily in AI models. Anthropic’s chatbot Claude is a competitor to OpenAI’s ChatGPT and Google’s Gemini.
Anthropic is in talks to raise up to $2 billion from Lightspeed and Google at a $60 billion valuation, CNBC has reported. Amazon has plowed roughly $8 billion into backing the startup.
Anthropic product chief Mike Krieger, who previously co-founded Instagram, said the hybrid approach is a way to simplify the chatbot process for customers. They can use multiple capabilities without needing to think about which is the best option.
“Models all have personalities, they’re all a bit different,” Krieger told CNBC, adding that it’s a “lot” to have consumers choose the model, or how long they want it to reason. “I would love for people, end users, not have to think about that very much at all.”
Krieger said users should be able to turn the hybrid option on or off for simplicity. They can give it a time “budget” based on what they’re working on. Anthropic will also roll out a tool for coding using agents on Monday.
The startup has had a few wins with product launches ahead of competitors. It was also the first to unveil a widely available “agent” capability late last year, which OpenAI soon followed. Krieger and Kaplan both said they expect competitors to move in this direction with hybrid models.
OpenAI CEO Sam Altman hinted at moving in a similar direction earlier in February. In a post on X, Altman said his AI company plans to do a “better job of sharing our intended roadmap, and a much better job simplifying our product offerings.”
“We hate the model picker as much as you do and want to return to magic unified intelligence,” Altman wrote.
WATCH: Anthropic CEO more confident than ever that we’re ‘very close’ to powerful AI capabilities
