San Francisco, Sept. 30 – Anthropic has unveiled its latest artificial intelligence system, Claude Sonnet 4.5, describing it as the “best coding model in the world” and a major step forward in building production-ready applications.
The new model, announced Tuesday, delivers stronger performance in software development, cybersecurity, finance, and research. It can run autonomously for up to 30 hours—four times longer than earlier versions—while maintaining focus on complex, multistep tasks.
“Claude Sonnet 4.5 is our most aligned frontier model yet,” the company said in a blog post, citing advances in reliability, safety, and defenses against prompt injection attacks. It outperforms previous Anthropic models on benchmarks such as SWE-Bench Verified and is designed to generate higher-quality code and follow instructions more consistently.
Pricing remains unchanged at $3 per million input tokens and $15 per million output tokens via the Claude API and chatbot.
The launch comes just weeks after rival OpenAI released GPT-5, underscoring the intensifying competition between the two San Francisco-based firms. Both are targeting enterprise clients adopting AI for software development—a trend dubbed “vibe coding,” where even non-technical users can build apps with AI-powered tools.
Backed by Amazon and valued at $183 billion, Anthropic has rapidly expanded since launching Claude in 2023, growing to more than 300,000 customers in two years. The company said its run-rate revenue has surged past $5 billion, up from about $1 billion earlier this year.
Anthropic is also accelerating global expansion with a new Asia office in Tokyo, over 100 hires in Dublin and London, and a research hub in Zurich. It plans to recruit country leads across India, Australia, New Zealand, South Korea, and Singapore.
