Introduction
Anthropic has unveiled Claude Sonnet 4.5, proclaimed as the world's best coding AI model. This breakthrough system represents a significant evolutionary leap, delivering superior capabilities in building complex agents, computer usage, and substantial improvements in mathematical reasoning.
Core Innovations of Claude Sonnet 4.5
Claude Sonnet 4.5 excels in the SWE-bench Verified evaluation, which measures real-world software coding abilities. The model maintains focus for over 30 hours on complex, multi-step tasks, demonstrating unprecedented operational endurance.
In the OSWorld benchmark, testing AI models on real-world computer tasks, Sonnet 4.5 achieves 61.4% success rate, surpassing the previous 42.2% record set by Sonnet 4 just four months ago.
Product Updates
- Claude Code: Added checkpoints to save progress and instantly roll back to previous states
- Refreshed terminal interface and native VS Code extension
- New context editing feature and memory tool for Claude API
- Code execution and file creation directly in conversations
- Claude for Chrome extension available to Max users
Claude Agent SDK: Developer Tools
Anthropic introduces the Claude Agent SDK, the same infrastructure powering Claude Code internally. This toolkit provides developers with essential tools for building custom AI agents, including solutions for memory management across long-running tasks and permission systems balancing autonomy with user control.
Advanced Safety and Alignment
Claude Sonnet 4.5 represents Anthropic's most aligned frontier model, showing significant improvements in reducing problematic behaviors like sycophancy, deception, power-seeking, and encouraging delusional thinking. The model releases under AI Safety Level 3 (ASL-3) protections, featuring classifiers to detect potentially dangerous inputs and outputs.
"We've made considerable progress on defending against prompt injection attacks, one of the most serious risks for users of these capabilities."
Anthropic Team, Research and Development
"Imagine with Claude" Experiment
Anthropic presents "Imagine with Claude," a temporary research preview where Claude generates software in real-time. The experiment demonstrates the model's creative capabilities, adapting and responding to user requests without predetermined functionality or prewritten code.
Availability and Pricing
Claude Sonnet 4.5 is available today via the Claude API using the identifier "claude-sonnet-4-5". Pricing remains unchanged from Claude Sonnet 4: $3/$15 per million tokens. The model serves as a drop-in replacement providing significantly improved performance at the same price point.
FAQ
What is Claude Sonnet 4.5 and why is it considered revolutionary?
Claude Sonnet 4.5 is Anthropic's new AI model, recognized as the world's best coding system. It offers superior capabilities in building complex agents and computer usage.
What are the main improvements of Claude Sonnet 4.5 over previous models?
The model shows substantial gains in mathematical reasoning, maintains focus for over 30 hours on complex tasks, and achieves 61.4% on the OSWorld benchmark.
How can I access Claude Sonnet 4.5 for my development projects?
You can use Claude Sonnet 4.5 through the Claude API with the identifier "claude-sonnet-4-5". Pricing remains $3/$15 per million tokens.
What is the Claude Agent SDK and how can it help developers?
The Claude Agent SDK is the infrastructure used to develop Claude Code, now available for building custom AI agents with advanced memory management and control systems.
What security measures does Claude Sonnet 4.5 implement?
The model operates under ASL-3 protections with classifiers detecting dangerous content, particularly related to CBRN weapons, and improved defense against prompt injection attacks.
Will the "Imagine with Claude" experiment be permanently available?
"Imagine with Claude" is a temporary research preview available to Max subscribers for five days at claude.ai/imagine.