Anthropic has just announced the release of a new AI model called Claude Opus 5.5 designed to address advanced coding, computer-use, and occupational knowledge work tasks. The newest model represents an expanding direction in the AI market, namely transitioning from systems that respond to prompts to AI agents capable of planning, tool use and executing multi-step tasks with minimal human input.
Overview. The new models: Claude Opus 5.5 has been enhanced to be faster and cheaper to operate than previous models and is especially appropriate for developers and business who need to use an AI agent for longer and more complex series of actions. Released 22 September 2026.
Claude Opus 5.5 Focuses on Agentic AI
In the new model, the most important thing to us was agentic work. Instead of a normal chatbot that pauses until the user provides it with an instruction, an AI agent can subdivide a high-level goal into lower-level tasks, communicate with tools and keep working on the goal.
For software engineers, it might involve writing code as a software engineer, maintaining codebases, and fixing bugs. “Anthropic’s strong Opus model of Opus 5.5 is the strongest Opus model for agentic coding, with a focus on long-running projects and larger codebases,” it explains.
The model provides a 1 million token context window and can generate 128,000 tokens of output. This enables it to handle a significant volume of information in long coding and knowledge-work sessions.
Lower Cost Could Be a Major Advantage
Cost has played a larger role in the AI model race than ever before as businesses are leveraging AI for production workloads rather than sporadically experimenting with it.
For Claude Opus 5.5, you are billed at a rate of $4 per million input tokens, $20 per million output tokens. Reading from cache costs $0.20 per million tokens. On average, operations have an estimated $0.40 savings in comparison to Opus 5.
Plus, the company says Opus 5.5 produces results over 30% faster than Opus 5. Businesses that operate AI agents in perpetuity may find that the combo of speed and token efficiency could influence the total cost to run automated workflows.
This is especially relevant to agentic applications as an agent may trigger thousands of model calls for a single job. The lower cost per call can therefore be significant when multiplied by thousands of jobs.
Claude Opus 5.5 Benchmark Performance
Open to business: technical skill in a new world The German start-up Anthropic announced strong results of its chatbot Claude Opus 5.5 on coding, computer-use and knowledge-work tests.
On Terminal-Bench 4.0, which challenges an AI system to carry out more involved command-line operations, Opus 5.5 achieved 66.4% in Anthropic’s published benchmark comparison. It also achieved 54.4% on Frontier Code and 57.8% on Cursor Bench.
To perform tasks more generally, the company achieved a score of 1,846 Elo on GDPval-AA v2.1, which assesses work in 44 occupations.
These numbers are derived from tests conducted by Anthropic and should not automatically be assumed to reflect real world performance on all tasks. The company itself observed that at this level of benchmark disparity, it does not always translate to a comparable difference in performance on tasks.
Computer Use Becomes More Important
Computing is another vital part of Claude Opus 5.5.
Beyond text, AI will also interact directly with digital environments. This might involve driving software, engaging with files, reading documents, interpreting diagrams or executing a number of commands through several applications.
Anthropic claims that Opus 5.5 can handle these multi-step problems better in tasks like computer usage and vision. The AI can perform tasks that require it to manipulate documents, screenshots, diagrams and other information.
For businesses, it could have implications for tools in document processing, research, finance, software and working inside the enterprise.
Safety Remains a Key Focus
As AI agents expand their ability to act autonomously, safety will also grow in importance in model development.
The company states Opus 5.5 is more resilient to prompt-injection attacks than Opus 5 and has been tested for coding, tool-use, computer-use and web-browsing cases. Anthropic states it has implemented extra protections for vital cybersecurity and biological tasks.
External organisations took part in assessing the model before release, according to Anthropic, but some capabilities will still be limited further.
What does Kurlan? Stanswer – Claude Opus 5.5 for the Fields of AI:
The release of Claude Opus 5.5 is just one example of a wider shift in AI. Chatbot responses are taking a backseat to agent-based systems.
Developers’ use of the model for coding and the vast context window may facilitate longer software projects. Businesses’ economy of lower token costs may enable permanent usage of AI-agents.
Using Anthropic: You can access Claude Opus 5.5 via Google’s platform, Amazon Web Services, Google Cloud and Microsoft Foundry. We have also announced the general availability of this model to all Claude Pro, Max, Team and Enterprise users.
By playing on speed, cost and capability at the same time, AI companies are going to build a world where the ability of a model to do valuable work efficiently can prove equally powerful. Claude Opus 5.5: Next up in the feature quality race.
