Anthropic Launches New AI Model Claude 3.5 Sonnet

Anthropic launched a new AI model on Thursday that it says It’s their “smartest model yet.”

The new model, Claude 3.5 Sonnet, is reportedly twice as fast as the Claude 3 Opus, the company’s previous best AI, and five times cheaper to operate. Following a trend set by its competitor OpenAI – which last month released the latest version of ChatGPT, GPT-4o – Claude 3.5 Sonnet is free for all users on the web and iOS to access. It was also made available to developers.

Claude 3.5 Sonnet is “now the smartest model in the world”, says Michael Gerstenhaber, the company’s product manager. “We are at the beginning of a Cambrian explosion in this industry,” Gerstenhaber told TIME ahead of the launch.

Looking at the pace of model releases, it’s easy to see why. The Claude 3.5 Sonnet arrives just three months after the release of Anthropic’s previous set of models, the Claude 3, and just under a year after the release of its ancestor, the Claude 2.

Unlike GPT-4o, the latest version of Claude cannot search the internet or generate image files. Your “intelligence” is measured by your performance on a series of benchmarks, which, although imperfect, show that it is currently at the front of the pack. But it is the qualitative aspects of the model, not its performance in benchmarks, that excite Gerstenhaber most. Early users of 3.5 Sonnet highlighted its intelligence and the humor it displayed in interactions, he said. This is likely the result of Anthropic’s efforts to give Claude a distinctive characterand have the model “demonstrate a genuine curiosity about the opinions and values ​​of the people they are talking to,” according to the company website.

See more information: Inside Anthropic, the AI ​​company betting that security can be a winning strategy

In addition to its improved image coding and transcription abilities, 3.5 Sonnet has a new feature that changes the way chatbots work – what Anthropic calls “artifacts.” When users ask Claude to write documents, generate code, or help with something like web design, he displays the requested content in a separate panel next to the chat. Ask to make changes to the content and you can see it updated in real time.

The artifacts offer a glimpse into how Anthropic is thinking about the future of its models, where they go beyond chatbots and act more like sandboxes where teams of people can collaborate on work, “with Claude serving as an on-demand teammate.” , according to the company’s press release. This suggests that future models may have more in common with collaborative software tools like Notion and Google Docs than with the chatbot interfaces that proliferate today.

Anthropic says the Sonnet 3.5 has undergone rigorous security testing. Concerns about the capabilities of current and future models are on the minds of virtually every major AI lab, most of which have published documents detailing what actions they will take based on how dangerous they consider their models to be — often called “scaling policies.” responsible.” (RSPs). Despite the increase in intelligence, the model is still considered at ‘AI Safety Level 2’ according to Anthropic’s PSR; on the same level as Claude 3 Opus was in March.

Sonnet 3.5 has been shared with the UK’s AI Safety Institute for pre-deployment testing ahead of its release. The results of these tests were also shared with the AI ​​Safety Institute in the United States, as part of a partnership between the two institutions.

Other information on the Sonnet 3.5’s performance in safety tests reveals that the model “does not pose a risk of catastrophic damage”, for example, increasing risks related to biological weapons or nuclear war. However, the documentation notes that Anthropic “observed an increase in capabilities in relevant areas of risk compared to Claude 3 Opus.”

Regarding privacy, Anthropic says it “does not train [their] generative models on user-submitted data unless a user gives us explicit permission to do so.” They also note that “to complete the Claude 3.5 model family,” updated versions of Opus and Haiku (respectively major and minor versions of Sonnet) will be released later this year.

-With additional reporting by Billy Perrigo/London