Anthropic debuts Claude 3.7 Sonnet with hybrid reasoning and launches Claude Code CLI
Anthropic has launched Claude 3.7 Sonnet, the industry's first hybrid frontier model combining instant responses with dynamic extended thinking budgets, alongside a repo-native terminal agent called Claude Code.
In a major architectural leap for generative artificial intelligence, Anthropic has officially unveiled Claude 3.7 Sonnet. The new release establishes a distinct architectural category as the market's first production hybrid reasoning model, integrating instantaneous response generation and deliberate, step-by-step reasoning ('Extended Thinking') into a single unified foundation.
Unlike previous paradigms—such as OpenAI's split between the fast GPT-4o and slower, dedicated o-series reasoning engines—Claude 3.7 Sonnet allows engineers to set an explicit thinking budget via API parameters, ranging from zero tokens for standard queries up to 128,000 tokens for exhaustive software engineering or mathematical investigations. During extended thinking, the model's internal thought progression is exposed to users, offering complete visibility into intermediate deduction steps.
Simultaneously, Anthropic introduced Claude Code, an agentic command-line interface engineered specifically for terminal-native software engineering. Claude Code navigates complex git repositories, edits source files across directory trees, executes unit test suites, resolves compiler diagnostics, and dispatches pull requests autonomously. Anthropic reported that Claude 3.7 Sonnet maintains standard pricing at $3 per million input tokens and $15 per million output tokens (including thinking tokens), with immediate availability across the Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI.
Why it matters
Developers no longer need to compromise between low-latency conversation and deep chain-of-thought problem solving; Claude 3.7 unifies both into a single model with user-controlled token budgets up to 128,000 thinking tokens.