Anthropic suddenly released the Claude Sonnet 4.5 version at night. This update not only achieved a comprehensive improvement in technical indicators, but also demonstrated an innovative direction for AI to generate software interfaces in real-time through the experimental feature "Imagine with Claude."
Claude Sonnet 4.5 has made significant breakthroughs in coding capabilities. The model achieved a score of 77.2% on the SWE-bench Verified benchmark test, surpassing the previous version Claude Opus 4.1's 74.5%, and was called the best coding model in the world by Anthropic. It performs excellently in logical reasoning, advanced mathematics, and multi-step coding tasks, and can independently run for over 30 hours to handle complex agent tasks.
Notably, despite the smaller model size compared to Opus 4.1, Sonnet 4.5 leads in most capability dimensions. The pricing strategy remains unchanged, with input at $3 per million tokens and output at $15 per million tokens. It is accessible through the Claude API, Amazon Bedrock, and Google Vertex AI.
In terms of security, the new version has optimized model alignment, reduced undesirable behaviors such as pleasing users or deception, and adopted the ASL-3 security framework to filter potentially dangerous content. The model has already been able to independently reconstruct the Claude.ai web application, completing over 3,000 tool calls in 5.5 hours, demonstrating near-production-level capabilities.
The core highlight of this release is the experimental feature "Imagine with Claude," which is available for a 5-day trial for Max subscribers. This temporary feature provides an interactive interface similar to a desktop environment. Users can input their requirements through natural language, and Claude Sonnet 4.5 will generate UI elements, functional logic, and interaction mechanisms in real-time.
Different from traditional software development, the Imagine feature does not rely on pre-set code or fixed templates, but dynamically generates complete applications based on user intent. For example, when a user requests to create a weather forecast application, the AI will instantly render the interface, implement functional logic, and handle data interactions. Users can also place the generated application in a desktop environment for use.
Along with the model release, Anthropic launched the Claude Agent SDK, allowing developers to build custom agents using internal infrastructure, supporting virtual machine access, memory management, and multi-agent collaboration. Claude Code added a checkpoint feature for instant progress rollback and integrated with VS Code and JetBrains development environments, as well as a Chrome extension for Max users.
The "Imagine with Claude" feature has sparked widespread discussions about AI-native operating systems. This real-time generation of interfaces challenges traditional software development paradigms, no longer relying on pre-written code, but allowing AI to dynamically evolve interfaces and logic based on user intent.
It should be noted that the Imagine feature is still in its early experimental stage. There are delays in button responses for complex UIs, which require further optimization. Anthropic stated that this move aims to explore the boundaries of agent capabilities and plans to expand it to more application scenarios in the future.
From a technological development perspective, Claude Sonnet 4.5 strengthens Anthropic's competitive position in the coding field, while the Imagine feature represents an exploration of a new human-computer interaction paradigm. This real-time generation of software interfaces could bring fundamental changes to future application development and user experience design.
However, it is necessary to maintain a rational understanding. The current experimental features are still far from a true AI-native operating system. Many issues such as stability, performance optimization, and security guarantees need to be gradually resolved in practical applications.
