An advanced AI language model that delivers superior coding and reasoning capabilities with more precise instruction following. Offers both near-instant responses and extended thinking modes for deeper reasoning tasks.
Claude Sonnet 4 is a Language Model from Anthropic that delivers state-of-the-art coding and reasoning capabilities with hybrid instant and extended thinking modes, with pricing starting at $3 per million input tokens and $15 per million output tokens. It is built for developers, engineering teams, and enterprises that need a balanced, high-throughput model for production workloads.
Released in May 2025 as part of the Claude 4 family alongside Claude Opus 4, Sonnet 4 represents a significant upgrade over Claude Sonnet 3.7, scoring 72.7% on SWE-bench Verified â one of the highest publicly reported scores for a frontier coding model at launch. The model introduces hybrid reasoning, meaning users can toggle between near-instant responses for routine queries and an extended thinking mode that lets the model deliberate for longer on complex problems. It also gains the ability to use tools (such as web search) during extended thinking, alternating between reasoning steps and tool calls to improve answers. Anthropic has tightened instruction-following behavior, reduced reward-hacking shortcuts by 65% versus Sonnet 3.7 on agentic coding tasks, and added parallel tool use plus improved memory when developers grant file access.
In practice, Claude Sonnet 4 powers GitHub Copilot's new coding agent, drives agentic workflows in Cursor, Windsurf, and Replit, and is available through the Claude API, Amazon Bedrock, and Google Cloud Vertex AI. Compared to GPT-4.1 and Gemini 2.5 Pro, Sonnet 4 is positioned as the most coding-capable mid-tier model â cheaper than Opus 4 (which runs $15/$75 per million tokens) but markedly stronger than competing mid-tier models on agentic and software-engineering benchmarks. Based on our analysis of 870+ AI tools, Sonnet 4 sits in the top tier of language models for production coding agents, IDE integrations, and long-horizon agentic tasks where reliability and instruction adherence matter more than raw chat capability.
Was this helpful?
Sonnet 4 can respond near-instantly for routine queries or switch into an extended thinking mode that allocates more compute for deliberation. Developers control this via a single API parameter, making it possible to escalate hard requests within a single agent loop without swapping models.
Unlike traditional models that finish reasoning before invoking tools, Sonnet 4 can interleave tool calls (such as web search or code execution) inside its extended thinking process. This produces more grounded answers on research-style questions and reduces hallucinations on factual claims.
Anthropic specifically tuned Sonnet 4 for long-running coding agents, achieving 72.7% on SWE-bench Verified and reducing reward-hacking shortcut behavior by 65% versus Sonnet 3.7. This is why GitHub selected it as the engine for Copilot's new coding agent and why it leads adoption inside Cursor, Windsurf, and Replit.
When granted file system access, Sonnet 4 can write notes to disk to maintain context across long sessions and call multiple tools in parallel rather than sequentially. This dramatically improves throughput and consistency on multi-hour agentic tasks like full-project refactors or research syntheses.
Sonnet 4 is available through the Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI at the same $3/$15 per million token pricing. Combined with prompt caching (up to 90% off cached inputs) and batch processing (50% off async workloads), this gives enterprises flexibility on procurement, compliance, and cost optimization.
$0
$20/month
$25/user/month (annual) or $30/user/month billed monthly, min 5 seats
$3 / $15 per million input/output tokens
Custom
Ready to get started with Claude Sonnet 4?
View Pricing Options âWe believe in transparent reviews. Here's what Claude Sonnet 4 doesn't handle well:
Weekly insights on the latest AI tools, features, and trends delivered to your inbox.
Claude Sonnet 4 launched in May 2025 as part of the Claude 4 family alongside Claude Opus 4, introducing hybrid extended thinking, tool use during reasoning, parallel tool calls, and a 65% reduction in reward-hacking shortcuts versus Sonnet 3.7. It scores 72.7% on SWE-bench Verified and powers GitHub Copilot's coding agent, Cursor, Windsurf, and Replit. As of April 2026, it remains widely deployed in production via the Anthropic API, Amazon Bedrock, and Google Cloud Vertex AI.
No reviews yet. Be the first to share your experience!
Get started with Claude Sonnet 4 and see if it's the right fit for your needs.
Get Started âTake our 60-second quiz to get personalized tool recommendations
Find Your Perfect AI Stack âExplore 20 ready-to-deploy AI agent templates for sales, support, dev, research, and operations.
Browse Agent Templates â