Anthropic has introduced Claude Sonnet 4.5, a new frontier model aimed squarely at software development. The company says the model delivers state-of-the-art results on coding benchmarks and is reliable enough to build "production-ready" applications rather than stopping at prototypes.
The launch adds another major entry to the fast-moving market for AI coding models, where developers, enterprises, and coding tools are weighing reliability, autonomy, price, and security claims with each new release.
What Anthropic says Claude Sonnet 4.5 can do
Claude Sonnet 4.5 is being positioned as Anthropic's strongest coding model to date. According to the company, it leads on several coding benchmarks, including SWE-Bench Verified.
Anthropic is not presenting the model only as a benchmark improvement. The larger claim is that Claude Sonnet 4.5 can handle longer and more complete software engineering work. That distinction matters because many AI coding tools can generate demos or partial features, while production work requires more sustained reasoning, coordination, and follow-through.
Anthropic AI researcher David Hershey told TechCrunch that benchmark scores do not fully capture the model's performance. He said he has seen Claude Sonnet 4.5 code autonomously for up to 30 hours during early trials with some enterprise customers.
In that period, Hershey said the model did more than write application code. He watched it build an application, stand up database services, purchase domain names, and perform a SOC 2 audit to make sure the product was secure.
Availability and pricing stay familiar
Claude Sonnet 4.5 will be available through both the Claude API and the Claude chatbot. For developers, Anthropic is keeping pricing the same as Claude Sonnet 4.
The price is $3 per million input tokens and $15 per million output tokens. The source describes one million input tokens as roughly 750,000 words, or more than the entire "Lord of the Rings" series.
Keeping the price steady is notable because the launch is framed as a step up in coding performance. For teams already using Claude Sonnet 4, the core pricing structure does not change with this release.
The model also arrives in a market where Anthropic's previous Claude models have already gained traction with developers and enterprises. The source notes that Apple and Meta reportedly use Claude AI models internally, and that Anthropic has built a significant business selling API access to AI coding applications including Cursor, Windsurf, and Replit.
Why coding agents are central to the launch
Alongside Claude Sonnet 4.5, Anthropic is launching the Claude Agent SDK. The company says this is the same infrastructure that powers Claude Code and that developers can use it to build their own agents.
That addition shows how much of the AI coding conversation has moved beyond single prompts and short code completions. The model is being launched with tooling meant for agent-style workflows, where software can take instructions, work across tasks, and interact with development infrastructure.
Statements from coding tool leaders in the source point in the same direction. Cursor CEO Michael Truell said Claude Sonnet 4.5 represents state-of-the-art coding performance, specifically on longer horizon tasks. Windsurf CEO Jeff Wang said Claude Sonnet 4.5 represents a "new generation of coding models."
Anthropic is also releasing a temporary research preview called "Imagine with Claude" for Max subscribers. The company says it shows the AI model generating software on the fly, responding to user requests in real time with no predetermined functionality or prewritten code.
Security and alignment claims are part of the pitch
Anthropic is also making safety and reliability claims around Claude Sonnet 4.5. The company says it is its most aligned frontier AI model yet, with lower rates of sycophancy and deception than previous models.
The company also says it has improved Claude's susceptibility to prompt injection attacks. For developers and enterprises, that claim is directly relevant to agentic coding workflows, because agents may interact with codebases, tools, instructions, and external content while completing tasks.
The source does not provide independent test results for those safety claims. But their inclusion in the launch message shows that Anthropic is treating coding performance and trustworthiness as connected parts of the same product story.
A fast-moving contest for AI coding leadership
Claude Sonnet 4.5 arrives at a time when the AI model market is moving quickly. The source notes that intense competition has made it common for companies to ship flagship models every few months.
That pace affects how long any one company can maintain a clear lead. Claude Sonnet 4.5 is launching less than two months after Anthropic's last AI model, Claude Opus 4.1.
Anthropic has become a favorite among developers and enterprises in the last year, largely because of strong performance on software engineering tasks. But the source also notes that OpenAI's GPT-5 has challenged Anthropic's dominance by outperforming Claude models on a variety of coding benchmarks.
The result is a market where each new model is judged quickly on practical value: whether it can build more complete software, work for longer, reduce failure modes, and justify its place inside developer tools. Claude Sonnet 4.5 is Anthropic's latest answer to that pressure.