Anthropic officially announced the latest “Claude 4” series of AI models on May 22, 20251. This release includes the top-tier “Claude Opus 4” and mid-tier “Claude Sonnet 4” models, with particular attention drawn to their significantly improved coding performance.
The Claude 4 series achieved remarkable scores on the software engineering benchmark SWE-bench, with Opus 4 scoring 72.5% and Sonnet 4 scoring 72.7%1. This benchmark measures the ability to solve real GitHub repository issues, serving as an important indicator of performance on realistic coding tasks.
Revolutionary New Features
Extended Thinking and Tool Use (Beta)
One of Claude 4’s most notable features is its “Extended Thinking and Tool Use” capability1. This allows the model to execute multiple tools in parallel and efficiently process more complex tasks. It also possesses the ability to work continuously for hours, making it suitable for large-scale projects and debugging work.
Memory Feature Implementation
Claude 4 can now create and maintain “memory files”1. This enables information retention between sessions and remembering long-term projects or user-specific settings. This feature is expected to significantly improve team work efficiency in enterprise environments.
Performance and Accuracy Improvements
More Accurate Instruction Execution
Claude 4 is reported to use 65% fewer “shortcuts” or “workarounds” compared to the previous generation1. This indicates improved ability to understand user intent more accurately and execute instructions as directed.
High Praise from Major Companies
Development tool companies have given high ratings, with Cursor calling it “state-of-the-art in coding” and Replit praising its “improved accuracy for complex changes”1. Furthermore, GitHub announced the adoption of Claude 4 for their new Copilot coding agent.
Pricing and Accessibility
Model-Specific Pricing Structure
Claude 4 pricing is as follows1:
- Claude Opus 4: Input $15 / Output $75 (per million tokens)
- Claude Sonnet 4: Input $3 / Output $15 (per million tokens)
Sonnet 4 is also available to free users, allowing more developers to experience Claude 4’s performance.
Available Platforms
Claude 4 is available on the following platforms1:
- Anthropic API
- Amazon Bedrock
- Google Cloud Vertex AI
- Pro, Max, Team, and Enterprise plans
Impact on Global Developers
The arrival of Claude 4 is expected to have a significant impact on the global developer community. Applications are particularly anticipated in the following areas:
-
Code Quality Improvement: The high SWE-bench score suggests high accuracy in actual bug fixes and feature additions.
-
Development Efficiency Enhancement: Parallel tool execution allows processing multiple tasks simultaneously, enabling shortened development cycles.
-
Multilingual Support: Anthropic focuses on multilingual support, with coding assistance in various languages expected.
Comparison with Competitors
While direct benchmark comparison data with GPT-4 or Gemini has not been published at this time, the over 72% performance on SWE-bench indicates top-tier performance as a coding-specialized AI. The advancing adoption in actual development environments particularly demonstrates its high practicality.
The evolution of AI development support tools, including future updates and competition with other models, remains an area to watch closely.
Sources
- Introducing Claude 4 - Anthropic Official Announcement (May 22, 2025)