Anthropic Releases Smartest Neural Network Claude 3.7 Sonnet – It’s Free and Outperforms DeepSeek R1 and OpenAI o3

Anthropic, one of OpenAI’s main competitors, has released Claude 3.7 Sonnet, its first “hybrid reasoning model.” The company says it can solve more complex problems than its predecessors, and outperforms them in areas such as math and coding.

Image source: Anthropic

OpenAI and others offer reasoning models separate from regular generative AI models. Anthropic decided to combine them into a single system to create a universal solution. The result is that the user can choose when the Claude 3.7 Sonnet model should respond normally, and when it should take a little longer to think about the answer. In standard mode, Claude 3.7 Sonnet is simply an improved version of the previous Claude 3.5 Sonnet with more recent data (its database includes information up to November 2024). In advanced reasoning mode, the AI ​​thinks for itself before answering, which improves performance on problems in math, physics, complex instructions, coding, and more.

Anthropic’s head of product research, Dianne Penn, told The Verge that the company wanted to make the model easier to use. “We fundamentally believe that reasoning is more of a feature of AI than a separate thing,” she said, noting that it doesn’t take Claude long to answer a question like “what time is it?” compared to a more complex query like “plan a two-week trip to Italy based on the weather in late March.”

In addition to the new model, Anthropic also released a “limited research preview” of its AI coding agent, called Claude Code. While Anthropic already offers AI coding tools like Cursor, the company bills the new Claude Code as “an active collaborator that can search and read code, edit files, write and run tests, commit and push code to GitHub, and use command-line tools.”

Anthropic also lets developers control how the model “thinks,” and even set a time limit for thinking. “Sometimes a developer just needs to say, ‘This question shouldn’t take more than 200 milliseconds to answer,’” says Michael Gerstenhaber, Anthropic’s vice president of product.

Penn claims that Claude 3.7 Sonnet is noticeably better than its competitors at “agent coding,” financial and legal tasks. According to an Anthropic spokeswoman, the company’s employees actively use the new model to create website designs, interactive games, and even spend up to 45 minutes on coding, “creating test cases and iteratively editing test cases.”

Penn also said the company tests its models to see if they can beat the old-school Pokémon video game by simulating controller button presses via an API. Claude 3.5 Sonnet was unable to escape Pallet Town early in the game, while version 3.7 was able to defeat several bosses.

The release of Claude 3.7 Sonnet shows that the AI ​​industry is moving toward offering a single model that can both respond quickly and think about complex problems, rather than several separate models. Something similar was recently discussed by OpenAI CEO Sam Altman.