Technology startup Anthropic introduced two new AI models Claude Opus 4 and Claude Sonnet 4 at its inaugural developer conference. These models, part of the Claude 4 family, are designed to analyze large datasets, execute long-horizon tasks, and perform complex operations, according to the company.
Both models are optimized for programming tasks, significantly enhancing code writing and editing efficiency. Paying users gain access to both Opus 4 and Sonnet 4, while free chatbot users can only use Sonnet 4. Via API on Amazon’s Bedrock platform and Google’s Vertex AI, Opus 4 is priced at $15/$75 per million tokens (input/output), and Sonnet 4 at $3/$15 per million tokens.
Tokens are the fundamental data units AI models process; one million tokens equate to approximately 750,000 words—about 163,000 words more than Tolstoy’s War and Peace.
Anthropic’s Claude 4 models align with the company’s ambitious revenue growth plans, aiming for $12 billion by 2027, up from a forecasted $2.2 billion this year. The startup has secured a $2.5 billion credit facility and attracted investments from Amazon and others to support the rising costs of frontier AI development.
The flagship Opus 4 model excels at maintaining “focused effort” through multiple workflow steps. Sonnet 4, a direct successor to Sonnet 3.7, offers improved performance in coding and mathematics, providing more accurate and faster responses.
Claude 4 models reduce “reward hacking,” where AI shortcuts task completion at the expense of correctness. However, while Opus 4 outperforms Google’s Gemini 2.5 Pro and OpenAI’s o3 and GPT-4.1 on the SWE-bench Verified coding benchmark, it does not surpass OpenAI’s o3 model on multimodal and advanced scientific question evaluations.
For security, Opus 4 is released with enhanced safeguards. Anthropic’s internal testing indicates the model could significantly increase capabilities for individuals with STEM backgrounds to acquire or develop chemical, biological, or nuclear weapons, categorized as their “ASL-3” security level.
Both models are “hybrid,” capable of near-instant responses and extended reasoning. When reasoning mode is enabled, they evaluate multiple solution paths before answering, presenting users with a concise summary of their thought process, while protecting proprietary competitive advantages.
Claude 4 models utilize multiple tools in parallel—such as search engines—and alternate between reasoning and tool usage to improve answer quality. They also retain facts in memory to increase task reliability and develop implicit knowledge over time.
To better serve programmers, Anthropic is enhancing Claude Code, a tool that integrates directly with developer environments like Microsoft VS Code, JetBrains, and GitHub. The newly introduced SDK allows developers to embed Claude Code into third-party applications, facilitating AI-powered coding assistance and automation.
Despite ongoing challenges—such as AI-generated code vulnerabilities and errors—the productivity gains drive rapid adoption among developers and companies.
Anthropic commits to more frequent model updates, promising faster delivery of breakthrough features. “We are shifting to a cadence of continuous improvements to keep customers at the cutting edge,” the company stated.
With the Claude 4 release, Anthropic continues its robust competition for leadership in the rapidly evolving AI landscape.
