Currently listed through these providers:
Model details
Claude Sonnet 4
Claude Sonnet 4 is a hybrid reasoning large language model positioned as a direct successor to Sonnet 3.7, built to deliver frontier performance for practical AI use cases rather than purely academic benchmarks. Anthropic designed it to balance capability and computational efficiency, with particular emphasis on coding, advanced reasoning, and instruction following. The model is meant to serve as a dependable engine for user-facing assistants and high-volume tasks alike, supporting extended thinking alongside tool use so it can alternate between reasoning and calling external tools like web search. It also introduced the ability to use tools in parallel, follow intricate instructions more reliably, and—when given access to local files—maintain continuity by extracting and saving key facts over time. Sonnet 4's practical strengths show up most clearly in agent-driven software development. On SWE-bench it reached 72.7%, a state-of-the-art result that reflects improved autonomous codebase navigation, lower error rates in multi-step agent workflows, and more reliable adherence to complex developer instructions. These gains make the model well suited to tasks ranging from routine code generation and refactoring to larger software engineering projects where sustained reasoning and careful tool use matter. Although later Sonnet releases have since extended the family, Sonnet 4 set the template for combining hybrid reasoning, parallel tool use, and memory-aware behavior into a single model aimed squarely at real-world deployment.
As part of Anthropic's Claude 4 generation, Sonnet 4 reflects the company's broader research direction toward models that pair deep reasoning with tight integration into agent ecosystems, an approach made visible by the simultaneous launch of Claude Code and the extended thinking with tool use beta. Its lineage traces through the Sonnet family that began with the hybrid reasoning breakthrough in Sonnet 3.7, with each iteration sharpening coding ability, instruction precision, and the capacity to handle long-running agentic workflows. The model is tuned for everyday production use rather than only research demos, prioritizing responsiveness alongside capability so it can fit into both interactive assistants and backend automation. In practical terms, Sonnet 4 is a strong fit for developers and teams building agentic coding assistants, automated code review pipelines, and AI copilots that need to navigate large codebases, chain multiple tools together, and follow detailed specifications without drifting. Its combination of state-of-the-art coding results, parallel tool execution, and improved instruction fidelity also makes it attractive for content generation, data analysis, and planning tasks where reliability matters as much as raw intelligence. While newer Sonnet variants have since pushed performance further on agents and long-horizon work, Sonnet 4 remains a meaningful milestone: the release that established hybrid reasoning with tool use as a practical default for production AI.
Quick Info
Powered by- Provider
- Vertex
- Model key
- claude-sonnet-4@20250514
- Release date
- May 22, 2025
- Last updated
- May 22, 2025
- Knowledge cutoff
- 2025-03-31
- AI SDK package
@ai-sdk/google-vertex/anthropic- Input modalities
- Output modalities
- Capabilities
Cost
- Input token cost
- $3.00
- Output token cost
- $15.00
Limits
- Output tokens
- 64,000 tokens
- Context window
- 200,000 tokens