Anthropic Launches Claude Sonnet 4.5: The Pinnacle of AI Coding Models
Anthropic's Claude Sonnet 4.5 launched as a leading AI coding model, boasting significant performance and safety improvements.
- • Claude Sonnet 4.5 is launched as the best coding model globally.
- • It achieved 61.4% on OSWorld benchmarks, outperforming competitors like OpenAI's GPT-5.
- • The model includes features like checkpoints, a refreshed interface, and a VS Code extension.
- • Safety enhancements have been introduced, reducing deceptive behaviors and prompt attacks.
Key details
Anthropic has officially launched Claude Sonnet 4.5, branding it as the "best coding model in the world". This new iteration boasts exceptional performance in coding tasks, effectively outperforming both its predecessor and several competing models in the market.
Claude Sonnet 4.5 is now accessible via the Claude API and chatbot, continuing with pricing similar to its predecessor at $3 per million input tokens and $15 per million output tokens. Notably, it demonstrated a significant leap in benchmark performance, scoring 61.4% on the OSWorld metric, an impressive increase from the 42.2% of Claude Sonnet 4. This enhancement positions it ahead of rivals such as OpenAI's GPT-5 and Google's Gemini 2.5 Pro, indicating robust advancements in both reasoning capabilities and reliability.
Enhancements include the introduction of checkpoints that allow users to save their work, an upgraded terminal interface, and a native VS Code extension. The model’s ability to operate autonomously for up to 30 hours significantly improves its utility for complex project management, setting a new standard in the AI coding landscape. Early insights reveal that developers have particularly praised its effectiveness in long-term coding tasks, with industry leaders acknowledging its potential.
In addition to performance enhancements, Anthropic emphasized safety improvements in Claude Sonnet 4.5, marking it as its most aligned AI model to date. With extensive training aimed at minimizing deceptive behavior and enhancing defenses against prompt injection attacks, it falls under AI Safety Level 3 protections.
The launch is accompanied by the release of the Claude Agent SDK, which allows developers to create tailored AI agents, and a temporary research preview named "Imagine with Claude" that showcases the model's software generation capabilities in real-time.
In summary, the launch of Claude Sonnet 4.5 underscores Anthropic's ongoing innovations and competitive edge in the fast-evolving AI sector, as awareness grows regarding the model's superior coding capabilities and safety features. The broader implications of this launch will continue to shape the dynamics within the AI coding model market as companies pursue advancements in collaboration and automation tools, amidst fierce competition from industry giants.