Anthropic has launched a new model, Claude Sonnet 4.5, which it says is performing better in coding tests. The company claims that Claude Sonnet 4.5 can create fully-fledged applications, not just prototypes. This could be a significant leap in reliability compared to previous AI models.
Anthropic has launched a new model, Claude Sonnet 4.5, which it says is performing better in coding tests. The company claims that Claude Sonnet 4.5 can create fully-fledged applications, not just prototypes. This could be a significant leap in reliability compared to previous AI models.
Claude Sonnet 4.5 will be available via the Claude API and the Claude chatbot. Developer pricing remains the same as for Claude Sonnet 4: $3 per million input tokens (roughly 750,000 words, or more than the entire Lord of the Rings series) and $15 per million output tokens, TechCrunch reports .
Anthropic says Claude Sonnet 4.5 is performing industry-leading in a number of coding tests, including SWE-Bench Verified. However, Anthropic AI researcher David Hershey told TechCrunch that it’s difficult to judge Claude Sonnet 4.5’s performance based on benchmarks alone.
Hershey says that in early testing with some enterprise customers, he saw Claude Sonnet 4.5 autonomously write code for up to 30 hours. During that time, he watched the AI model not only build the application, but also deploy database services, purchase domain names, and conduct SOC 2 audits to ensure the product was secure.
In a statement to TechCrunch, Michael Truell, CEO of Cursor, noted that Claude Sonnet 4.5 demonstrates the highest performance in programming, especially in tasks with long-term perspectives. Windsurf CEO Jeff Wang said in a statement that Claude Sonnet 4.5 is “a new generation of models for writing code.”
According to Anthropic, Claude Sonnet 4.5 is their most balanced model to date, exhibiting less spoofing and deception than previous versions. The company says it has also improved Claude's resistance to prompt injection attacks.
Along with the launch of Claude Sonnet 4.5, Anthropic is also launching the Claude Agent SDK. The company says it’s the same infrastructure that powers Claude Code, and it can be used to help developers build their own agents.
Anthropic is also introducing a temporary research version called Imagine with Claude for Max subscribers, which demonstrates how the AI model generates software “on the fly.” The company says the model will respond to user requests in real time, without predefined functionality or pre-written code.
