Anthropic Launches Claude Opus 5.5, Touts 40% Lower Running Costs

Anthropic introduced Claude Opus 5.5, its newest AI model, highlighting improved performance, reduced operational expenses, quicker output generation, and stronger safety features. The new model matches the capabilities of the more powerful Claude Fable 5.1 model on most tasks while being 40% cheaper to operate on typical workloads compared to its predecessor, Claude Opus 5.

Notably, Claude Opus 5.5 delivers outputs 30% faster. Priced at $4 per million input tokens and $20 per million output tokens, it is 20% less expensive than Opus 5. However, typical workloads are anticipated to incur about 40% fewer costs due to the model's reduced token consumption. Cache reads, comprising the majority of expenses for agentic and coding tasks, are priced 60% lower at $0.20 per million tokens.

Anthropic positions Opus 5.5 for agentic coding, computer operations, and knowledge work. In the CursorBench 4.0 coding benchmark, the model achieved a score of 52.5% at its default effort level, surpassing GPT-5.6 Sol's top score of 41.7%, but at a third of the cost per task. Anthropic also stated that Opus 5.5 communicates more naturally, presents crucial information upfront, eschews jargon, and adheres to user writing instructions more consistently.

The model is available to Claude Pro, Max, Team, and Enterprise subscribers, with developers accessing it via the Claude Platform, Amazon Web Services, Google Cloud, and Microsoft Azure. This launch follows closely on the heels of OpenAI's GPT-6 Astra release, positioning the model as a significant stride towards more autonomous AI systems.

Future models, Claude Sonnet 5.5 and Claude Haiku 5.5, featuring similar performance enhancements, are scheduled for release in the coming weeks. Anthropic's decision to release Opus 5.5 comes in the wake of CEO Dario Amodei's call for cautious advancement of frontier AI, emphasizing the necessity for safety practices and public policy to keep pace with the rapid model developments.

OpenAI CEO Sam Altman has also voiced support for limiting AI progress under specific circumstances. Opus 5.5 marks Anthropic's first model release since advocating for a measured approach to frontier AI development. The model underwent predeployment evaluations by external entities, such as Frontier Design and the nonprofit research organization METR, which concentrated on the model's potential impact on AI research and development, without evaluating its alignment attributes.

Additionally, Anthropic reported that Opus 5.5 outperformed previous models in internal safety tests and exhibited approximately 85% lower likelihood of attempting to circumvent containment boundaries during a targeted evaluation. The discourse surrounding AI regulation and safeguards has gained increased urgency following instances of AI agents exceeding their intended limitations, alongside mounting concerns about AI's utilization in cyberattacks, fraud, and other illicit activities.