Artificial intelligence company Anthropic has released Opus 5.5, its latest and most capable Claude model, saying the system delivers state-of-the-art performance in coding and knowledge work while costing less to operate.
Opus is the highest-capability tier in Anthropic’s three-model Claude lineup, alongside Sonnet and Haiku.
According to Anthropic, Opus 5.5 outperforms the larger Fable model on several benchmarks and also completed a number of informal tasks that Fable was unable to finish.
Lower costs and faster performance
Anthropic has also reduced the price of using its flagship model.
Output tokens for Opus 5.5 cost $20 per million tokens, down from $25 for its predecessor. The company said other pricing metrics have also been reduced.
The new model is also faster, with Anthropic attributing the improvement to lower computing requirements for serving the system.
The changes could make advanced AI capabilities more accessible to developers and businesses that use large models for coding, analysis and other knowledge-intensive tasks.
Changes to how Opus communicates
Anthropic said Opus 5.5 has also been trained to communicate more effectively.
The model is less likely to rely on technical jargon and is more likely to place important information near the beginning of its responses.
The company said the changes are intended to make interactions with the model clearer and more useful, particularly during complex knowledge-work tasks.
Sonnet and Haiku updates coming
The release comes only two months after Anthropic introduced Opus 5 on July 24.
Anthropic said updated versions of its other Claude models, Sonnet 5.5 and Haiku 5.5, are expected to arrive “in the coming weeks” with similar performance improvements.
The company has not provided a specific release date for either model.
Safety measures remain a focus
Anthropic said Opus 5.5 has capabilities comparable to its Fable model in areas including biology and cybersecurity.
As a result, the company said the model is subject to safeguards designed for highly capable AI systems.
These measures place restrictions on certain uses, including discovering exploits in compiled software and assisting with the development of biological weapons.
Anthropic said Opus 5.5 underwent alignment testing and pre-release evaluations involving external organisations including METR and Frontier Design.
Anthropic signals more cautious approach
The launch is also notable because it is Anthropic CEO Dario Amodei’s first major model release since calling for a more measured pace in the development of frontier AI systems.
Amodei has argued that AI capability development should proceed at a pace that allows safety and alignment efforts to keep up.
Anthropic said its safety training for Opus 5.5 was broadly similar to previous models, while more advanced training, evaluation, security and monitoring systems are being developed for future releases.
“As AI becomes more capable, public policy should play a larger role in making sure the systems people rely on are safe,” Anthropic said.
The company added that it has begun building infrastructure to support those efforts and expects to provide more details in the future.
