Anthropic has introduced Claude 4, the latest version of its assistant family and a direct competitor to OpenAI’s operator-style agents. Claude 4 ships with a refined “computer use” feature that lets the model view a screen, move a cursor, and interact with desktop applications under user supervision.
Agentic Workflows Take Center Stage
The upgrade is aimed at knowledge workers who need help with repetitive multi-step tasks. Early demos show Claude booking flights, reconciling spreadsheets, and drafting research summaries by pulling information from several browser tabs. Anthropic stresses that the agent pauses for human confirmation on purchases, password fields, and outbound emails.
Beyond simple automation, Claude 4 can chain actions across applications. For example, it can open a CRM, identify overdue deals, draft personalized follow-up messages, and schedule them for review. The user remains in control, with each step visible and editable before execution.
Safer by Design
Claude 4 also debuts Anthropic’s latest constitutional-classifier stack, which is designed to refuse requests that could enable fraud, malware, or misinformation at scale. The company has published an expanded model card covering risk evaluations, frontier-red-team findings, and policy compliance scores. Claude 4 Sonnet remains available for lighter latency-sensitive use cases.
Availability and Pricing
Anthropic is rolling out Claude 4 through its API, the Claude web app, and Amazon Bedrock. Pricing follows the existing token-based model, though agentic sessions that include screenshots and desktop actions consume more input tokens. Google Cloud customers can access Claude 4 through Vertex AI as part of an expanded partnership.
Industry Impact
Industry watchers view this announcement as another sign that the artificial intelligence market is shifting from raw capability demonstrations toward production-ready features. Buyers are increasingly focused on total cost of ownership, data governance, vendor transparency, and long-term support. The move also pressures competitors to respond quickly, which should accelerate innovation and drive more flexible pricing across the market. For end users, the practical result is likely to be better tools, clearer licensing terms, and stronger safety guardrails as the industry matures through 2025 and 2026. Enterprises that move early may capture meaningful workflow efficiencies before these capabilities become table stakes.