Anthropic has launched Claude Opus 5.5, the first model in its new Claude 5.5 family, focused on coding, computer use, and other complex professional tasks. The company says the model can deliver performance comparable to Fable 5.1 across many workloads while costing significantly less to operate than the previous Opus 5 model.
Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. https://t.co/Q9C2VKQ79f
— Claude (@claudeai) September 22, 2026
The release is aimed particularly at users running demanding AI workloads, including software development and agent-based tasks. Anthropic is also introducing lower token prices, faster output and higher usage limits for some subscribers, making the new model a potentially more practical option for sustained workloads.
Claude Opus 5.5 targets demanding professional workloads
Anthropic says Claude Opus 5.5 leads its internal benchmark results in areas including agentic coding, computer use and knowledge work. On the GDPval-AA v2.1 benchmark, which evaluates performance across 44 occupations, the model scored 1,846, compared with 1,735 for Fable 5.1.
The company has also highlighted the model’s ability to handle large software engineering projects. During early testing, Opus 5.5 reportedly completed a migration involving 680,000 lines of code in less than a day. In another internal test, it converted HAProxy from C to Rust in 9.5 hours, compared with 12 hours for Fable 5.1.
These examples are intended to demonstrate the model’s ability to work through lengthy, multi-step tasks rather than generate individual pieces of code. Agentic coding systems can inspect existing projects, make changes across multiple files, and complete a sequence of related development tasks with limited human intervention.
Anthropic is also positioning Opus 5.5 as an improvement for general knowledge work. The company says the model produces clearer writing than Opus 5, responding to earlier feedback that some answers could be difficult to follow. Anthropic says output generation is also more than 30% faster than the previous model.
Lower prices target high-volume AI use
One of the main changes with Claude Opus 5.5 is its pricing. Anthropic has set the cost at $4 per million input tokens and $20 per million output tokens. Both prices represent a 20% reduction compared with Opus 5.
The company has made a larger reduction to cached input pricing. Cache reads now cost $0.20 per million tokens, a 60% decrease. Cached information can be particularly important for agentic and coding workloads, where the same project context may need to be processed repeatedly during a long task.
The lower prices are designed to reduce the cost of running large and repeated AI workloads. For developers and organisations using models extensively, token costs can become a significant part of an AI application’s operating expenses, particularly when agents perform multiple actions during a single task.
Anthropic is also increasing five-hour usage limits for Pro, Max and Team subscribers. The change gives subscribers more room to use the model during longer sessions before reaching their usage limits. Combined with the lower token prices, the changes could make Opus 5.5 more suitable for users who rely on AI throughout a working day.
The company is presenting the price reductions alongside the performance improvements rather than positioning Opus 5.5 simply as a cheaper model. Its stated aim is to provide higher levels of capability while reducing the cost of running demanding tasks.
Safety remains a focus as Opus 5.5 expands availability
Anthropic has also changed the model’s safety behaviour, with particular attention to autonomous AI agents. The company says Opus 5.5 is less likely than Opus 5 to take actions that are difficult to reverse and has improved resistance to prompt injection attacks.
The focus comes as AI systems are increasingly being given the ability to interact with software, websites and other digital systems. Such capabilities can make agents more useful for development and research, but they can also introduce additional risks when an AI system encounters malicious instructions or takes unintended actions.
Anthropic is applying additional restrictions to some sensitive areas. Certain cybersecurity requests are being routed to Opus 4.8, while some biology-related requests are handled by Opus 5. Organisations that require broader access to biology capabilities can apply through Anthropic’s Life Sciences Verification Program.
Anthropic has also acknowledged a limitation in its testing of Opus 5.5. The company found that the model can sometimes appear to recognise when it is being evaluated. Anthropic says this behaviour makes it more difficult to predict how the model will behave in real-world situations based solely on benchmark testing.
Claude Opus 5.5 is now available through Anthropic as well as Amazon Web Services, Google Cloud and Microsoft Azure. Its launch gives developers and organisations another option for running advanced coding, computer-use and knowledge-work applications. At the same time, lower pricing and expanded subscriber limits are intended to support broader use.
