At its inaugural developer conference on Thursday, Anthropic unveiled two advanced AI models—Claude Opus 4 and Claude Sonnet 4—designed to significantly enhance their capabilities in data analysis, long-horizon task execution, and handling complex actions. As part of the newly introduced Claude 4 family, these latest models stand out for their capacity to excel in coding tasks, positioning them ideally for software development work, according to the company.
Claude Opus 4, the more powerful of the two, offers sustained and focused reasoning across multi-step workflows, making it particularly adept for intricate tasks that require prolonged cognitive processing. Claude Sonnet 4, meanwhile, serves as an upgraded, “drop-in replacement” for its predecessor Sonnet 3.7, providing substantial improvements in coding accuracy, mathematical reasoning, and task adherence.
Availability for these models differs slightly; while both paid subscribers and users of Anthropic’s free chatbot will have access to Claude Sonnet 4, premium access to Claude Opus 4 will remain exclusive to paid subscribers. Pricing via Anthropic’s API, offered on Amazon’s Bedrock platform and Google’s Vertex AI, will be set at $15 per million tokens input ($75 per million tokens output) for Opus 4. Sonnet 4 will be considerably cheaper, at $3 per million tokens input ($15 per million tokens output). Tokens are fundamental units of data used by AI models—with a million tokens roughly equivalent to about 750,000 words, slightly more extensive than Tolstoy’s “War and Peace.”
Anthropic’s announcement comes amid a highly competitive landscape as major rivals such as OpenAI and Google continue to release increasingly capable AI products. Despite Anthropic previously launching the Claude Sonnet 3.7 model and a coding-focused tool called Claude Code earlier this year, accelerated developments from competitors have frequently shifted the benchmarks for industry leadership.
Nevertheless, Anthropic emphasizes rigorous improvements in its latest offerings. According to its internal testing, Claude Opus 4 now significantly surpasses models like Google’s Gemini 2.5 Pro and OpenAI’s latest offerings on certain specialized coding benchmarks. However, Anthropic acknowledges that Opus 4 doesn’t lead in every metric: it falls short compared to OpenAI’s o3 model in other evaluations measuring multimodal capability as well as advanced scientific reasoning.
The newest line of models also includes improved safeguards against unwanted behavior. Anthropic pointed to Opus 4’s substantial increase in potential security risks if misused—including its heightened capability to facilitate development of harmful chemical, biological, or nuclear materials. As a result, Anthropic intends to apply stronger usage restrictions and enhanced cybersecurity protocols.
Another major advancement shared by the company is the adoption of “hybrid” structures, allowing the models to deliver nearly immediate responses for simpler queries, while enabling slower, more thoughtful analysis for complex reasoning problems. Users activating reasoning mode will see concise summaries illustrating the logic steps used by the model, rather than full transparency of its underlying decision-making process—a concession Anthropic says helps shield its competitive advantages.
Further bolstering capabilities for developers, Anthropic is rolling out additional integrations and upgrades to its software tool Claude Code. Developers can now directly link Claude Code to popular integrated development environments (IDEs) such as Microsoft’s VS Code, JetBrains, and GitHub. These upgraded integrations allow Anthropic’s AI to automate complex coding tasks, respond effectively to reviewer suggestions, and fix detected errors within existing codebases.
Aware of persistent challenges that AI-powered coding assistants face—with vulnerabilities and logic errors still relatively prevalent—Anthropic is doubling down on frequent iterative updates and continuous refinement. The company states clearly that achieving robust, flawless execution remains a work in progress, and a heavier update frequency will ensure users continuously benefit from state-of-the-art improvements.
Anthropic, founded by a group of ex-OpenAI researchers, is also firmly focused on building revenue momentum. The startup recently secured substantial financing, including a $2.5 billion credit line and significant investments from Amazon, intended to offset escalating development costs. It anticipates a significant increase in earnings—aiming for up to $12 billion by 2027, compared to a projected $2.2 billion in revenue this year.
