Anthropic launches Claude 3.7 Sonnet, a hybrid reasoning model, and unveils Claude Code
Claude 3.7 Sonnet can answer instantly or reason deeply on a configurable budget at unchanged prices, alongside a Claude Code preview that runs in the terminal.
Anthropic has launched Claude 3.7 Sonnet, a hybrid reasoning model, and introduced the research preview of Claude Code, a programming tool that runs in the terminal.
The novelty lies in choosing how to answer within the same model: instant responses for simple questions, or an extended thinking mode with a compute budget configured by the user for hard problems, while pricing stays unchanged. Claude Code is a coding agent in the terminal, meaning a tool that takes requests in natural language and works directly on source code rather than only suggesting line by line.
For enterprises, a hybrid model reduces the need to maintain a fast model and a dedicated reasoning model in parallel, which simplifies integration and quality assessment. A coding agent bears directly on the software development process: code review, bug fixing, writing tests and maintaining legacy systems, while also raising the question of how work should be divided between engineering teams and tools.
One point to note is that extended thinking mode consumes additional processing time and tokens, so costs can rise on heavy tasks even though list prices are unchanged. An agent with read and write access to source code also needs scoped access limits, logging, and a human review step before changes reach the main branch.
Source: anthropic.com
