Anthropic launched its new model, Opus 5.5, on September 22. The company states that this version achieves its highest internal test performance to date in coding and knowledge-based tasks, outperforming the larger Fable model across multiple benchmarks.
Price reduced, faster performance
Compared to the previous generation, the cost of using Opus 5.5 continues to decline. Anthropic states that the price for output tokens has decreased from $25 per million tokens to $20, with other billing metrics also reduced to varying degrees.
The company also stated that the new model runs faster and has reduced overall computational requirements. In addition to performance and cost, Opus 5.5 has adjusted its response style to use less jargon and prioritize placing key information at the beginning of replies.
Update again within two months
This release comes just about two months after Opus 5 launched. Anthropic stated in its announcement that Sonnet 5.5 and Haiku 5.5 will also be released in the coming weeks, with similar performance improvements expected.
From a product timeline perspective, Anthropic continues to rapidly iterate its models, addressing performance, cost, and deployment efficiency in each update simultaneously. The competitive focus is shifting from single benchmark scores to real-world performance.
Adopt Fable-level security measures
Anthropic states that Opus 5.5 has approached Mythos in biological and cybersecurity capabilities, and therefore is subject to the same level of safety restrictions as Fable. These measures primarily target high-risk uses, including identifying vulnerabilities in compiled programs and assisting in the development of identifiable biological weapons.
The company also noted that this is the first model release since CEO Dario Amodei publicly supported slowing the pace of advancing frontier capabilities. Anthropic stated that Opus 5.5 follows the existing safety training process and has undergone pre-release evaluations by external organizations such as METR and Frontier Design.