Anthropic Launches Claude Opus 5.5 with Enhanced Safety, Lower Cost
Anthropic has launched its new AI model, Claude Opus 5.5. This model reportedly delivers performance comparable to the company's Fable 5.1 system. It also costs 40 per cent less to operate than its direct predecessor. The introduction of this model occurs amidst ongoing discussions concerning artificial intelligence risks within the technology sector.
Dario Amodei, the chief executive of Anthropic, recently urged the global AI community to reduce the development speed of new capabilities. This call aimed to allow more time for addressing various safety concerns associated with artificial intelligence. The release of Opus 5.5 follows these statements, with the company stating specific safeguards were integrated. Independent AI safety research organisations, Frontier Design and METR, conducted external testing of the model prior to its public release, according to the organisation. These measures align with precautions previously reserved for Anthropic's most advanced systems.
Anthropic asserts that Opus 5.5 outperformed OpenAI's rival GPT-5.6 Sol model on a crucial software development benchmark. This assessment also indicated that running Opus 5.5 required approximately one third of the operational cost compared to the competitor. And the company has specified the pricing structure for its new offering. Users will pay USD 4 per million input tokens and USD 20 per million output tokens for Opus 5.5. This pricing represents a 20 per cent reduction compared to the previous Opus 5 model.
The Claude Opus 5.5 model is accessible across several major cloud computing platforms. These include Amazon Web Services, Google Cloud, and Microsoft Azure, indicating broad availability for developers and businesses. The company additionally claims the model achieved improved results in its internal safety evaluations. It was about 85 per cent less likely than its predecessors, Opus 5 or Mythos 5.1, to attempt bypassing containment boundaries during a dedicated evaluation process.
Anthropic plans to release Claude Sonnet 5.5 and Claude Haiku 5.5 in the weeks ahead. These forthcoming models are expected to incorporate many of the same enhancements observed in Opus 5.5. Such improvements are projected to encompass performance, operational speed, and overall safety features. This staggered release strategy suggests a continued focus on iterative upgrades across the Claude product line.
The organisation is also expanding its current programmes. These initiatives provide vetted cybersecurity and life sciences researchers with broader access to the advanced AI model. But access aims to facilitate further research and understanding of the model's capabilities and potential applications. This expansion shows the company's commitment to collaborative safety and development within the artificial intelligence ecosystem, as outlined by its leadership.
According to Reuters, these developments occur within a period of intensified discussion regarding AI risks. The company's chief executive had previously urged a slowing of development to address safety concerns. This indicates a continuing dynamic where advances in artificial intelligence models are accompanied by internal and external appeals for caution.
Anthropic launched Claude Opus 5.5, claiming comparable performance to Fable 5.1 at 40 per cent lower cost than its predecessor.
The model underwent external testing by Frontier Design and METR, including safeguards for advanced systems.
Opus 5.5 reportedly outperformed OpenAI's GPT-5.6 Sol on a software development benchmark and costs USD 4 per million input tokens, USD 20 per million output tokens.
Future releases include Claude Sonnet 5.5 and Claude Haiku 5.5, with similar performance, speed, and safety improvements expected.
Anthropic expands programmes providing vetted cybersecurity and life sciences researchers with broader model access.
Source: Reuters
