Anthropic Unveils Claude Opus 5.5: Enhanced Performance at Lower Costs
Anthropic has officially launched Claude Opus 5.5, the latest addition to its AI model lineup, aimed at enhancing coding and knowledge work capabilities. This new model not only boasts improved performance metrics but also significantly reduces operational costs, making it an attractive option for users in various professional fields. The company claims that Opus 5.5 performs at a level comparable to the previous Claude Fable 5.1, yet it costs 40% less to operate under typical workloads than its predecessor, Opus 5.
The pricing structure for Opus 5.5 has been strategically set at $4 per million input tokens and $20 per million output tokens, which represents a 20% decrease from the costs associated with Opus 5. Additionally, the efficiency of cached reads has improved dramatically, dropping 60% to just $0.20 per million tokens, while output generation speed has increased by over 30%.
In terms of performance, Anthropic has reported notable advancements in agentic coding capabilities. The Opus 5.5 model achieved a score of 66.4% on the Terminal Bench 4.0, surpassing the previous Opus 5, which scored 52.3%, and even outpacing GPT 6 Astra at 57.9%. Furthermore, it performed well on CursorBench 4.0 and FrontierCode 1.1, scoring 57.8% and 54.4%, respectively.
The design of Opus 5.5 particularly caters to long-running software tasks. Early testers have reported impressive results, including the completion of a 680,000 line code migration in less than a day and a 200,000 line codebase audit and repair in under three hours. These capabilities underscore the model's efficiency and effectiveness in handling complex coding tasks.
Advancements in Knowledge Work Performance
Anthropic has also highlighted the model's improvements in knowledge work performance. Opus 5.5 achieved an Elo score of 1,846 on the GDPval AA 2.1 benchmark, outperforming Fable 5.1 and Opus 5, which scored 1,735 and 1,708, respectively. This benchmark evaluates professional tasks across 44 different occupations, indicating the model's versatility and applicability in various fields.
Why It Matters
This release marks a significant milestone for Anthropic, being the first since CEO Dario Amodei emphasized the importance of pacing the development of frontier AI technologies. The Opus 5.5 model underwent rigorous external evaluations by METR and Frontier Design, and Anthropic claims it has achieved its best results on its internal behavioral alignment audits to date. Additionally, the new model introduces enhanced safeguards for cybersecurity, biology, and model distillation, with a reported 85% reduction in attempts to cross containment boundaries compared to Opus 5 and Claude Mythos 5.1.
Claude Opus 5.5 is now available across Anthropic’s platforms and major cloud services, including Amazon Web Services, Google Cloud, and Microsoft Azure. Following this launch, Anthropic is expected to release additional models, Claude Sonnet 5.5 and Claude Haiku 5.5, in the coming weeks.




