Anthropic introduced its latest frontier model, Claude Sonnet 4.5, on Monday, touting it as delivering top-tier results on coding benchmarks. According to the company, Claude Sonnet 4.5 can create “production-ready” software, not just prototypes, marking a significant improvement in reliability over earlier AI models.
Claude Sonnet 4.5 is accessible through both the Claude API and the Claude chatbot. Developer pricing remains unchanged from Claude Sonnet 4: $3 for every million input tokens (approximately 750,000 words, which exceeds the length of the entire “Lord of the Rings” trilogy) and $15 per million output tokens.
Over the past year, Anthropic’s AI models have become a preferred choice for developers and businesses, largely due to their strong capabilities in software engineering. Reports indicate that Apple and Meta utilize Claude AI models internally, and Anthropic has built a robust business by providing API access to coding tools like Cursor, Windsurf, and Replit. However, OpenAI’s GPT-5 has recently emerged as a strong competitor, surpassing Claude models on several coding benchmarks.
Anthropic asserts that Claude Sonnet 4.5 achieves leading results on multiple coding benchmarks, such as SWE-Bench Verified. Yet, Anthropic AI researcher David Hershey told TechCrunch that benchmarks alone do not fully reflect Claude Sonnet 4.5’s capabilities.
Image Credits:Anthropic Hershey shared that, during initial trials with enterprise clients, Claude Sonnet 4.5 was able to autonomously write code for as long as 30 hours. During these sessions, the AI not only developed an application but also set up database services, registered domain names, and conducted a SOC 2 audit to ensure product security.
In a statement to TechCrunch, Cursor CEO Michael Truell described Claude Sonnet 4.5 as delivering cutting-edge coding performance, especially for complex, long-term tasks. Windsurf CEO Jeff Wang called Claude Sonnet 4.5 a “new generation of coding models.”
Anthropic further claims that Claude Sonnet 4.5 is its most aligned advanced AI model so far, with reduced tendencies toward sycophancy and deception compared to earlier versions. The company also notes improvements in Claude’s resistance to prompt injection attacks.
Coinciding with the release of Claude Sonnet 4.5, Anthropic is unveiling the Claude Agent SDK. The company states that this is the same system behind Claude Code and is intended to help developers create their own agents.
Additionally, Anthropic is offering a limited-time research preview called “Imagine with Claude” for Max subscribers, which demonstrates the AI generating software in real time. The company says the model will answer user prompts instantly, without relying on preset features or prewritten code.
The fast-paced rivalry in the AI sector has led to frequent releases of flagship models. Claude Sonnet 4.5 arrives less than two months after Anthropic’s previous model, Claude Opus 4.1. Such accelerated development cycles make it challenging for any company to maintain a significant advantage for long.
