Anthropic, one of OpenAI’s main competitors, has introduced Claude 3.7 Sonnet, its first “hybrid reasoning model.” The company claims it can solve more complex problems than its predecessors and excels in areas such as mathematics and coding.
Claude 3.7 Sonnet is now available in the Claude apps and the web version of the chatbot, including for free users. It can also be accessed via Anthropic’s API, Amazon Bedrock, and Google Cloud’s Vertex AI. Despite the improvements, the pricing remains unchanged from its predecessor, Claude 3.5 Sonnet—$3 per million input tokens and $15 per million output tokens.
A Universal AI Solution
Unlike OpenAI and other companies, which offer separate reasoning models, Anthropic has integrated reasoning capabilities into one system. Users can decide when Claude 3.7 Sonnet should respond normally and when it should take additional time to generate a more thoughtful answer.
In standard mode, the model functions as an improved version of Claude 3.5 Sonnet, with updated training data that includes information up to November 2024. In advanced reasoning mode, it can analyze and process complex instructions, improving performance in areas like math, physics, and coding.
Dianne Penn, Anthropic’s head of product research, explained to The Verge that the company wanted to make reasoning a built-in feature of AI rather than a separate function. She pointed out that Claude 3.7 Sonnet can quickly answer simple questions like “What time is it?” but requires more processing power for complex queries such as planning a two-week trip to Italy based on March weather conditions.
New AI Coding Agent and Future Developments
Alongside Claude 3.7 Sonnet, Anthropic introduced a “limited research preview” of its AI coding agent, Claude Code. While the company already provides AI-powered coding tools like Cursor, Claude Code is described as a more interactive assistant. It can search and read code, edit files, write and run tests, push code to GitHub, and use command-line tools.
Anthropic also allows developers to control the model’s reasoning process, including setting time limits for responses, adds NIXSolutions. “Sometimes, you just have to tell the AI that this answer needs to be generated in 200 milliseconds,” said Michael Gerstenhaber, Anthropic’s vice president of products.
According to Penn, Claude 3.7 Sonnet performs significantly better than its competitors in agent coding, financial analysis, and legal tasks. Employees at Anthropic have already used the model to design websites, develop interactive games, and spend up to 45 minutes coding and refining test cases.
The company has even experimented with using its models to complete classic Pokémon video games by simulating controller button presses via an API. While Claude 3.5 Sonnet struggled to escape Pallet Town at the start of the game, Claude 3.7 managed to defeat several bosses.
The release of Claude 3.7 Sonnet reflects the AI industry’s shift toward unified models that can both deliver quick responses and handle complex reasoning. OpenAI CEO Sam Altman has also discussed similar advancements, indicating that hybrid AI models could become the standard.
We’ll keep you updated as more developments emerge in AI technology.