"Claude Code" Research Preview Launches — an Autonomous Coding Tool for the Terminal
On February 24, 2025, alongside the announcement of its new model “Claude 3.7 Sonnet,” Anthropic released “Claude Code,” an autonomous coding tool that runs from the command line, as a limited research preview. Marketed as a “proactive collaborator,” it lives in a developer’s terminal and handles everything from searching and understanding code to editing, running tests, and committing to GitHub.
Details
- Capabilities: Handles code search and comprehension, file editing, writing and running tests, committing and pushing to GitHub, and running command-line tools — all in one continuous workflow
- Availability: Released as a “limited research preview,” rolling out progressively to developers who signed up according to the documentation — not yet a general release, but still at an experimental stage
- Early results: In internal testing, some tasks that normally take 45 minutes or more of manual work were completed in a single run. Particularly effective for test-driven development, complex bug investigation, and large-scale refactoring
- Claude 3.7 Sonnet, announced simultaneously: A hybrid model that can switch between a normal response mode and an “extended thinking” mode that takes more time to reason. Via the API, developers can specify a thinking token budget of up to 128,000 tokens
- Planned improvements: Anthropic said it planned to improve tool reliability, support for long-running commands, and in-app display, based on user feedback
What happened next
About three months after the research preview, in May 2025, Claude Code moved to general availability (GA) alongside the announcement of Claude 4. Within roughly half a year after that, its annual recurring revenue (ARR) reached the $1B mark, and it grew rapidly into the core of Anthropic’s agentic product strategy.