Anthropic Launches Claude Opus 4.5 AI Model
Anthropic has released Claude Opus 4.5, marking a significant advancement in artificial intelligence capabilities. The new model demonstrates superior performance in software engineering, autonomous agents, and computer use tasks compared to previous versions and competing systems.
The model achieves state-of-the-art results on real-world software engineering benchmarks. In internal testing, Claude Opus 4.5 scored higher than any human candidate on a difficult performance engineering exam within a two-hour time limit. The company notes this raises important questions about how artificial intelligence will transform engineering as a profession.
Pricing has been reduced to $5 per million input tokens and $25 per million output tokens, making advanced capabilities more accessible to individual users, development teams, and enterprise customers. The model is available through the company’s applications, API, and three major cloud platforms using the identifier claude-opus-4-5-20251101.
Early testing by enterprise customers reveals substantial improvements in efficiency. The model uses significantly fewer tokens to complete tasks while maintaining or improving output quality. Users report that Claude Opus 4.5 handles complex multi-system debugging, code refactoring, and long-duration autonomous tasks more effectively than its predecessor, Sonnet 4.5.
The release includes a new effort parameter that allows developers to control whether the model prioritizes speed and cost efficiency or maximum capability. At medium effort settings, the model matches Sonnet 4.5 performance while using 76 percent fewer output tokens. At maximum effort, it exceeds Sonnet 4.5 by 4.3 percentage points while still using 48 percent fewer tokens.
Safety improvements accompany the performance gains. Anthropic states that Claude Opus 4.5 demonstrates the strongest resistance to prompt injection attacks among current frontier models. These attacks attempt to smuggle deceptive instructions into prompts to manipulate model behavior. The company developed testing protocols in collaboration with security research firm Gray Swan.
The model shows enhanced creative problem-solving abilities. During benchmark testing designed to measure agent capabilities, Claude Opus 4.5 identified an unexpected but legitimate solution to a customer service scenario that the benchmark designers had not anticipated. The model suggested upgrading a ticket class first, then modifying flight details, working within policy constraints in an inventive way.
Platform updates include improved context management and memory capabilities that boost performance on complex agentic tasks by nearly 15 percentage points in company evaluations. The model can effectively coordinate teams of sub-agents, enabling sophisticated multi-agent system construction.
Consumer applications receive notable upgrades. Claude Code now features Plan Mode, which asks clarifying questions before creating an editable plan file and executing tasks. The desktop application supports multiple parallel sessions for simultaneous work on different projects. Long conversations in the Claude app no longer reach context limits, as the system automatically summarizes earlier exchanges.
Claude for Chrome, which enables task handling across browser tabs, becomes available to all Max subscribers. Claude for Excel, announced in October, expands from limited preview to all Max, Team, and Enterprise users. Testing shows 20 percent accuracy improvement and 15 percent efficiency gains in financial modeling and Excel automation tasks.
Usage limits have been adjusted for Opus 4.5 access. Opus-specific caps have been removed for Claude and Claude Code users. Max and Team Premium subscribers receive increased overall limits roughly equivalent to their previous Sonnet token allocations.
The model demonstrates improvements across multiple capability areas beyond software engineering. Enhanced vision, reasoning, and mathematics skills make Claude Opus 4.5 competitive across numerous evaluation benchmarks. The company reports that certain complex tasks, such as advanced 3D visualizations and long-context storytelling, show particular improvement.
Development platform enhancements include effort control, context compaction, and advanced tool use features. These additions enable longer operation duration, expanded capabilities, and reduced need for human intervention during autonomous task execution.
