Anthropic Launches Claude Opus 5.5 AI Model with 40% Lower Costs
Anthropic has unveiled Claude Opus 5.5, the latest in its flagship AI lineup that offers increased performance, faster responses, better safeguards, and up to 40% less expensive to run. The model is priced 20% below the Claude Opus 5 and targeted knowledge work, computer use and agentic coding.
The newest flagship AI model from Anthropic, Claude Opus 5.5, boasts better performance, reduced running costs, quicker output, and improved safeguards, among other things. Anthropic claims that, for most tasks, Claude Opus 5.5 matches the performance of their more robust Claude Fable 5.1 model. On average, this new model is 40% more cost-effective to run than Claude Opus 5 was.
Additionally, the new model speeds up production by more than 30%. Claude Opus 5.5 is priced 20% lower than Opus 5, with $4 per million input tokens and $20 per million output tokens. Yet, because of the new model's reduced token usage, Anthropic predicts that normal workloads will be 40% cheaper. The majority of the costs associated with agentic and coding workloads, according to Anthropic, are cache reads, which are priced 60% lower at $0.20 per Mn tokens.
Loaded Features of Opus 5.5
Anthropic is getting Opus 5.5 ready for knowledge work, computer usage, and agentic coding. The model outperformed GPT-5.6 Sol, which achieved a top score of 41.7% on the CursorBench 4.0 coding benchmark, with a score of 52.5% at its default effort setting. Nevertheless, the company claims that their pricing is approximately one-third as much per work. Additionally, the business claimed that compared to earlier versions, Opus 5.5 communicates more organically. More than that, it continuously follows the writing instructions of its users, eliminates jargon, and prioritises key information.
Users with Claude Pro, Max, Team, or Enterprise subscriptions have access to the model. It is accessible to developers via the Claude Platform, AWS, Google Cloud, and Azure. This launch is scheduled for a little over three weeks following OpenAI's GPT-6 Astra launch. This puts the model in a prominent position to pave the way for increasingly independent AI systems. Astra is capable of navigating the web, interacting with programs, working with spreadsheets and documents, writing and testing code, and executing workflows with multiple steps.
Opus 5.5 Far Safer and Secure AI Model: Anthropic
Frontier Design and the nonprofit research organisation METR were among the external organisations that evaluated Opus 5.5 before deployment. According to Anthropic, they didn't test the model's alignment qualities in their early review; instead, they looked at how the model could affect AI research and development. In a separate assessment, Anthropic stated that Opus 5.5 performed better than earlier systems in internal safety testing.
Further, the firm claims that it is approximately 85% less likely to try to escape containment limits compared to Opus 5 or Mythos 5.1. Incidents involving AI agents working outside their intended boundaries and growing concerns over AI being exploited for cyberattacks, fraud, and other criminal actions have intensified the discussion over frontier AI legislation and safeguards. Dario Amodei, the CEO of Anthropic, issued a call to AI companies to advance the frontier earlier this month. This leads one to the conclusion that public policy and safety procedures must evolve to accommodate the ever-improving capabilities of models. Even OpenAI CEO Sam Altman has said that there are times when it's appropriate to pause AI research and development.
