Anthropic Releases Claude 3.7 Sonnet: World's First Hybrid Reasoning Model
Anthropic introduces Claude 3.7 Sonnet featuring adjustable thinking budget parameters, allowing developers to balance instant latency with deep chain-of-thought mathematical reasoning.
Source: Anthropic Official Research Blog ↗ Published: 2026-08-20
Anthropic has officially released Claude 3.7 Sonnet, introducing a revolutionary hybrid architecture that allows developers to dynamically scale reasoning computational expenditure via an API thinking budget parameter.
Figure 1: Claude 3.7 Sonnet neural network architecture supporting hybrid reasoning modes.
Key Capabilities & Benchmarks
- SWE-bench Verified Score: Achieves a record 70.3% resolution rate on real-world GitHub issues.
- Adjustable Thinking Budget: Developers can specify exact max token limits for chain-of-thought processing.
- Claude Code CLI Integration: Serves as the primary default engine driving Anthropic's new terminal agent.