Anthropic Launches Claude Opus 5.5 as First Model in New 5.5 Family
By Suad Seferi ·
Anthropic has released Claude Opus 5.5, the first model in its new Claude 5.5 family, with the company focusing this update on lower operating costs, faster responses and stronger performance in coding and knowledge work. The model became available on September 22 across Claude and Anthropic's developer platforms. Lower costs and stronger performanceAnthropic says Opus 5.5 delivers performance comparable to its higher-end Claude Fable 5.1 model on most tasks, while costing considerably less to operate than the previous Opus 5. API pricing is set at $4 per million input tokens and $20 per million output tokens. Anthropic estimates that typical workloads will cost around 40% less than with Opus 5, partly because the new model can complete tasks using fewer tokens. The company also says output generation is more than 30% faster. Those changes make the release particularly relevant for developers and companies using Claude inside software development, internal tools and agent-based workflows, where the cost of repeated model calls can quickly become more important than small differences on individual benchmarks. Coding and computer use remain a priority Anthropic is positioning Opus 5.5 primarily as a model for complex work rather than everyday lightweight requests. Benchmark results and evaluation caveatsAccording to the company's internal evaluations, the model improves on Opus 5 across agentic coding, computer use and knowledge-work tasks. Anthropic also says it performs better than both Opus 5 and Fable 5.1 across most of the benchmarks included in its release testing. As with benchmarks from any model developer, those results are produced under controlled testing conditions and will need to be compared with independent evaluations and real-world use. Anthropic has also worked on how Claude communicates. The company says Opus 5.5 is designed to put important information earlier in its responses, follow writing instructions more consistently and avoid some of the unnecessarily elaborate writing users reported with earlier versions. That may sound like a relatively small change compared with benchmark improvements, but it becomes important when AI models are used across long coding sessions, document creation and business workflows where users need predictable output rather than simply longer answers. Safety becomes part of the release story Safety testing and deployment safeguardsOpus 5.5 is also Anthropic's first major model release since the company renewed its focus on keeping safety measures ahead of increasingly capable models. The model underwent testing by external evaluators including METR and Frontier Design before release. Anthropic says its latest alignment testing also showed fewer cases in which the model attempted to work around restrictions placed on it. The company reported an 85% reduction in this behaviour compared with Opus 5 and Mythos 5.1. The release arrives during a period of growing debate across the AI industry about how quickly frontier models should be deployed and how companies should evaluate increasingly autonomous systems before giving them access to tools, computers and external services. For Anthropic, the answer appears to be continuing model development while putting more emphasis on external evaluation and safeguards around sensitive areas such as cybersecurity and biology. More Claude 5.5 models are coming Opus 5.5 is not expected to remain the only member of the new model generation. Future models and regional implicationsAnthropic has indicated that Claude Sonnet 5.5 and Claude Haiku 5.5 will follow, extending the 5.5 generation across the company's different performance and price tiers. That will make the Sonnet release particularly important to watch. Sonnet has traditionally occupied the middle of Anthropic's lineup and is often the more practical option for developers who need capable models at higher volumes. For companies and developers in the Balkans, the broader significance of the 5.5 generation may therefore be less about another benchmark leader and more about how much frontier-level AI can be delivered for each euro spent on inference. Opus 5.5 starts that shift at the top of Anthropic's product line. The next test will come when the same efficiency improvements reach Sonnet and Haiku.