Claude Opus 5.5: Features, Pricing, Coding and AI Agent Capabilities
Claude Opus 5.5 is Anthropic's latest Opus model, introduced as part of the Claude 5.5 family. Its focus extends beyond conversational answers to complex coding, agentic workflows, computer use, professional knowledge work, and more efficient model usage.
Anthropic introduced Claude Opus 5.5 on September 22, 2026. The company presents it as a model built for demanding work that can continue across multiple steps rather than stopping after a single response.
That distinction is important because many advanced AI workflows involve more than generating text. A coding agent may need to inspect a large codebase, make changes, test those changes, correct errors, and continue working. A professional knowledge task may involve documents, spreadsheets, charts, research, and several rounds of analysis.
Claude Opus 5.5 is designed around these longer workflows. Anthropic highlights improvements in coding, computer use, agentic work, communication, and safety while also emphasizing lower token costs and reduced token consumption on typical workloads.
What Is Claude Opus 5.5?
Claude Opus 5.5 is Anthropic's latest Opus model and the first model in its Claude 5.5 family, according to Anthropic. It is intended for complex tasks that benefit from sustained reasoning, tool use, computer interaction, and extended execution.
The model is particularly focused on software engineering and agentic work, but Anthropic also positions it for professional knowledge tasks, financial analysis, computer use, and other workflows where the model may need to operate through several stages.
Claude Opus 5.5 Pricing
Anthropic lists Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens. Cache reads are priced at $0.20 per million tokens, while cache writes cost $5 per million tokens.
These prices are lower than the corresponding prices Anthropic lists for Opus 5. Anthropic says Opus 5.5 also uses fewer tokens to complete typical tasks, producing an approximately 40% lower cost on typical workloads billed by token compared with Opus 5.
| Pricing | Claude Opus 5.5 | Claude Opus 5 |
|---|---|---|
| Input | $4 / 1M tokens | $5 / 1M tokens |
| Output | $20 / 1M tokens | $25 / 1M tokens |
| Cache reads | $0.20 / 1M tokens | $0.50 / 1M tokens |
| Cache writes | $5 / 1M tokens | $6.25 / 1M tokens |
Anthropic also describes a faster mode for Opus 5.5 in Claude Code and on the Claude Platform. The company says this mode can reach up to 2.5 times the speed, with pricing of $8 per million input tokens and $40 per million output tokens.
Coding Is One of the Main Priorities
Claude Opus 5.5 is designed for software work that extends beyond generating an isolated code snippet. Anthropic highlights large codebase migrations, audits, debugging, refactoring, feature development, and code review.
Anthropic describes examples involving substantial codebases. One early tester used Opus 5.5 for a 680,000-line code migration that was completed in less than a day. Another used it to audit and fix a 200,000-line codebase in under three hours, compared with more than 20 hours for Opus 5 in that test.
Anthropic also reports that Opus 5.5 succeeded 39 out of 40 times in a task involving reducing load times across every page of a web application. According to Anthropic, Opus 5 sometimes made smaller improvements while also changing application behavior in the comparison.
The broader point is that the model is intended to work across a software system rather than treating each coding request as an isolated prompt.
Claude Opus 5.5 Benchmark Results
Anthropic reports improvements for Opus 5.5 across several evaluations covering coding, computer use, knowledge work, visual chart recognition, and multidisciplinary reasoning.
For agentic coding, Anthropic reports a 66.4% result on Terminal-Bench 4.0 for Opus 5.5, compared with 52.3% for Opus 5. On FrontierCode v1.1, Anthropic reports 54.4% for Opus 5.5 versus 48.0% for Opus 5.
On CursorBench 4.0, Anthropic reports 57.8% for Opus 5.5 compared with 46.6% for Opus 5. For professional knowledge work, the company reports a GDPval-AA v2.1 score of 1846 for Opus 5.5 versus 1708 for Opus 5.
For computer use, Anthropic reports an 81.8% partial score on OSWorld 2.0 for Opus 5.5 compared with 74.0% for Opus 5.
These figures describe particular evaluation setups. Anthropic also notes that benchmark differences are becoming less useful as a standalone guide to real-world performance, so individual results should be considered alongside the actual workflow being evaluated.
Agentic Work and Computer Use
A major theme of Claude Opus 5.5 is agentic work. Anthropic describes the model as capable of orchestrating complex multi-tool tasks, coordinating subagents, maintaining memory across sessions, and continuing longer-running work with limited oversight.
Computer use extends this idea beyond text. The model is designed to work with screenshots, documents, charts, and diagrams and to perform multi-step tasks across applications.
This changes the practical role of an AI model. Instead of producing information that a person must always carry into another application manually, an agent can participate directly in parts of the workflow. The actual reliability of such a workflow still depends on the tools, permissions, instructions, safeguards, and evaluation surrounding the model.
Communication and Usability
Anthropic says Opus 5.5 communicates more naturally than previous models. The company specifically highlights putting important information first, reducing unnecessary jargon, and following writing rules more reliably.
For extended AI workflows, communication quality has a practical role. People need to understand what a model has produced and identify where additional review is required. Clearer output can make that process easier.
Anthropic also connects communication quality with safety, arguing that clearer communication can make it easier for people to understand and supervise AI work.
Safety and Model Safeguards
Anthropic says Opus 5.5 achieved its strongest result to date on the company's automated behavioral audit, which evaluates Claude across thousands of simulated scenarios.
According to Anthropic, the model is less likely than recent models to take hard-to-reverse actions or operate outside defined boundaries. The company also reports stronger resistance to prompt injection compared with Opus 5.
Anthropic says its evaluation program now includes longer tasks, impossible tasks, and scenarios modeled on real incidents. The company nevertheless notes that Opus 5.5 still has limitations.
Because Anthropic considers Opus 5.5 capable in areas such as biology and cybersecurity, the company says additional safeguards are used for sensitive research applications. Some access therefore involves verification programs rather than unrestricted availability.
Where Claude Opus 5.5 Fits
The model's published positioning centers on tasks where sustained work and tool interaction are important. The main areas highlighted by Anthropic include:
- Software engineering: large migrations, debugging, audits, refactoring, code review, and feature development.
- Agentic workflows: multi-tool tasks, subagent coordination, memory, and longer-running execution.
- Professional knowledge work: complex documents, analysis, spreadsheets, presentations, and related tasks.
- Financial analysis: financial documents, models, charts, reporting, and related professional workflows.
- Computer use: working with visual information and performing tasks across applications.
The common element is complexity. Claude Opus 5.5 is positioned for work that may require context, tool use, verification, and several stages of execution rather than a single short answer.
Key Ideas
- Claude Opus 5.5 is built for complex workflows: coding, agents, computer use, and professional knowledge work are central to its positioning.
- Efficiency is a major part of the release: Anthropic reports lower prices and lower token usage compared with Opus 5.
- Coding remains a major focus: Anthropic highlights large-scale migrations, codebase audits, debugging, and other software engineering tasks.
- Computer interaction is central: Opus 5.5 is designed to work with visual information and multi-step application tasks.
- Communication has been improved: Anthropic emphasizes clearer output, better prioritization, and stronger adherence to writing instructions.
- Benchmarks need context: evaluation results represent specific tests and should be interpreted within their stated methodology.
Claude Opus 5.5 Moves AI Further Into Extended Workflows
The significance of Claude Opus 5.5 is not limited to another increase in benchmark scores. Anthropic is presenting the model as a system for work that can continue through multiple stages, use tools, interact with computers, and handle complex software and knowledge tasks.
At the same time, the release puts considerable emphasis on efficiency, communication, and safeguards. For people evaluating the model, those characteristics are part of the story alongside its reported performance.