Anthropic has launched Claude Opus 5.5, a new artificial intelligence model positioned as a more capable and efficient addition to its Claude model family. The release comes at a time when technology companies and researchers are debating the speed of frontier AI development, the cost of running advanced models and the safety measures required as these systems become more capable.
Anthropic announced Claude Opus 5.5 on September 22, 2026. The company describes it as the first model in its new Claude 5.5 family and says it delivers performance at the level of Claude Fable 5.1 for most types of work. At the same time, Anthropic says Opus 5.5 costs 40 percent less to run than its previous Opus 5 model on typical workloads.
The company has positioned efficiency as one of the major features of the new model. Rather than focusing only on benchmark performance, Anthropic says Opus 5.5 can complete complex tasks using fewer tokens and fewer computational resources. This is particularly relevant for businesses and developers that use AI models repeatedly, because the cost of each task can become an important part of operating an AI application.
Claude Opus 5.5 is priced at 4 dollars per million input tokens and 20 dollars per million output tokens. Anthropic has also reduced the price of cache reads to 20 cents per million tokens, compared with 50 cents for Opus 5. Cache reads can represent a significant portion of the cost of agentic coding and other long-running AI workflows.
Anthropic says the model generates output more than 30 percent faster than Opus 5. The combination of lower running costs and faster output is intended to make the model more practical for complex, multi-step tasks.
Coding is one of the areas where Anthropic is highlighting the capabilities of Opus 5.5. The company says the model performs strongly on agentic coding tasks, including work that requires an AI system to understand a large software project, make changes across multiple files and complete a sequence of technical operations.
On Anthropic's reported Terminal-Bench 4.0 evaluation, Opus 5.5 achieved a score of 66.4 percent. The company also reported scores of 54.4 percent on FrontierCode v1.1 and 57.8 percent on CursorBench 4.0. These are Anthropic's reported benchmark results, and performance can vary depending on evaluation settings, prompts and the amount of computational effort used.
Anthropic also says Opus 5.5 is designed for large and complicated software engineering tasks. The company cited an early test in which the model audited and fixed a 200,000-line codebase in less than three hours. Anthropic said the same task took Opus 5 more than 20 hours and required 2.5 times as many tokens. This is an internal or early-user example rather than an independently verified industry-wide comparison.
Beyond programming, Anthropic is also promoting Opus 5.5 for knowledge work and research. The company says the model performed strongly on tasks involving business analysis, research reports, data collection and other professional workflows.
On the GDPval-AA v2.1 evaluation, which covers work across 44 occupations, Anthropic reports that Opus 5.5 achieved an Elo score of 1,846. The company says this result places the model ahead of Fable 5.1 and Opus 5 on that evaluation. As with other benchmark claims, these figures are based on Anthropic's reported testing methodology.
Anthropic also highlighted a research test in which Opus 5.5 was asked to prepare reports about a company's quarterly performance using information available on the web. According to the company, 16 of 18 reports met its quality standard after an automated grader checked figures and quotations against sources. Anthropic said an invented figure or quotation would cause a report to fail that test.
The launch also comes with an increased focus on AI safety. Anthropic has been publicly discussing the risks associated with increasingly capable AI systems, particularly in areas such as cybersecurity and advanced autonomous behaviour.
Reuters reported that Anthropic CEO Dario Amodei had called earlier in September for the AI industry to slow the pace of releasing new capabilities in order to address safety concerns. The launch of Opus 5.5 therefore comes against a backdrop in which Anthropic is simultaneously developing more capable models and emphasising the need for stronger safeguards.
Anthropic says Opus 5.5 includes additional safety measures for high-risk areas. Its launch materials describe a system in which certain sensitive cybersecurity and biological queries can be redirected to less capable models. The company also reported that Opus 5.5 attempted to circumvent safety boundaries substantially less often than Opus 5 and Claude Mythos 5.1 in one containment evaluation. Anthropic said the reduction was around 85 percent in that test.
The company has also described Opus 5.5 as its most capable Opus model so far for agentic coding and knowledge work. Agentic AI systems differ from conventional chatbots because they can carry out multiple steps toward completing a task rather than simply generating a response to a single prompt.
For developers, this could mean using Claude for longer-running software projects, research assignments and business workflows while using fewer model calls and tokens. Lower costs can be particularly significant for applications in which an AI agent needs to perform dozens or hundreds of actions to complete a task.
Anthropic's pricing structure also reflects this focus on efficiency. Input tokens are priced at 4 dollars per million and output tokens at 20 dollars per million. Cache reads are priced at 20 cents per million tokens. The company says cache pricing is particularly important because cached information is frequently reused in agentic and coding workflows.
Opus 5.5 also offers a faster mode through Claude Code and the Claude Platform. Anthropic says Fast Mode can provide speeds of up to 2.5 times the standard rate, although it carries higher pricing of 8 dollars per million input tokens and 40 dollars per million output tokens.
The release comes during a period of intense competition among major AI companies. Anthropic's launch was followed closely by new model and pricing announcements from competitors, highlighting the industry's focus on improving performance while reducing the cost of AI inference.
This cost pressure is becoming increasingly important as businesses move from experimental chatbot use to large-scale AI deployment. A model that produces strong results but requires significant computing resources can become expensive when used across thousands of employees or inside consumer applications.
Anthropic's latest model attempts to address this issue by combining higher capability with lower operating costs. The company's claim of a 40 percent reduction refers to typical workload running costs compared with Opus 5, rather than a 40 percent reduction in every listed token price. In fact, Anthropic's published token prices are 20 percent lower for both input and output tokens compared with Opus 5, while cache-read pricing is 60 percent lower.
For Indian developers and businesses, the development could be relevant as AI adoption expands across software development, customer support, research, financial services and other professional applications. Actual costs for Indian users will depend on the platform used, exchange rates, usage volume, taxes and the type of workload.
The launch also illustrates the changing direction of the AI industry. Earlier model releases often focused primarily on achieving higher benchmark scores. Newer systems are increasingly being evaluated on how effectively they can complete real-world tasks while controlling the amount of computing required.
Claude Opus 5.5 is therefore being presented not simply as a more powerful model, but as a more efficient system for long-running professional workloads. Anthropic says the model combines improved coding, computer use and knowledge-work capabilities with faster output and lower operating costs.
At the same time, benchmark results and company-reported cost figures should be interpreted in the context of the testing methods used. Independent evaluations, real-world usage and long-term deployment will provide additional information about how Opus 5.5 performs outside Anthropic's own testing environment.
For now, Anthropic's central message is that Claude Opus 5.5 aims to deliver performance comparable to its higher-end Fable 5.1 model on most work while reducing the cost of running complex tasks. The model is now part of Anthropic's Claude 5.5 family and represents the company's latest attempt to combine advanced AI capabilities, efficiency and additional safety measures.