Anthropic Launches Claude Opus 5.5 With 40% Lower Costs

Overview

Anthropic has officially launched Claude Opus 5.5, marking the debut of its new Claude 5.5 family of models. Released on September 22, 2026, the new flagship represents a significant shift in the company’s strategy, prioritizing efficiency and accessibility without compromising on frontier-level capability.

The model is designed to perform at the level of Claude Fable 5.1 on most tasks while reducing operational costs by 40% compared to its predecessor, Opus 5. This release follows Anthropic’s recent public call for pacing the frontier in AI development, signaling a deliberate move toward more responsible deployment of highly capable systems.

Performance and Efficiency Gains

Claude Opus 5.5 delivers substantial improvements in computational efficiency. The model requires less compute to serve than Opus 5, a reduction that is directly reflected in its pricing structure. On typical workloads, running the model costs 40% less, with input tokens priced at $4 and output tokens at $20 per million—a 20% decrease from the previous generation.

"Opus 5.5 is the new leading model, and early testers saw large jumps in performance on their most complex work."

One of the most dramatic demonstrations of this capability involved a user who completed a massive 680,000-line code migration in less than a day. Similar projects using previous iterations of the model would have required an engineering team several weeks to execute. The model also excels at identifying and correcting software inefficiencies; in internal tests, it successfully reduced load times across every page of a web app 39 out of 40 times, whereas Opus 5 often made changes that inadvertently altered application behavior.

Benchmark data further underscores these capabilities. Opus 5.5 leads in agentic coding, achieving a 66.4% score on Terminal-Bench 4.0 and 57.8% on CursorBench 4.0. In the realm of computer use, it scored 81.8% on OSWorld 2.0, outperforming both Opus 5 and other major competitors.

Safety and Alignment

As the first model released since Anthropic called for paced frontier development, safety remains a central pillar of the Opus 5.5 launch. The model underwent rigorous testing by external evaluators, including Frontier Design and METR, prior to its public release.

On Anthropic’s automated behavioral audit—described as the most comprehensive alignment test the company runs—Opus 5.5 achieved the strongest performance of any model tested to date. The evaluation covers thousands of simulated scenarios, assessing the model’s resistance to prompt injection and its likelihood to take irreversible actions or exceed its assigned boundaries. Early results indicate that Opus 5.5 is significantly more resistant to prompt injection than Opus 5.

"It is much less likely than recent models to take hard-to-reverse actions or act outside the boundaries it’s been given."

To mitigate risks associated with high-stakes applications, Anthropic has aligned the safeguards on Opus 5.5 with those deployed for Claude Fable 5.1. This is particularly relevant for capabilities in biology and cybersecurity, where the model performs comparably to Mythos 5.1. Vetted organizations can now apply to the Life Sciences Verification Program to utilize Opus 5.5 for biological research. Additionally, access to the Cyber Verification Program will expand in the coming weeks, allowing verified practitioners to use the model for security work.

Communication and User Experience

Beyond raw metrics, Anthropic has focused heavily on the naturalness of Opus 5.5’s communication. Early testers noted that the model’s writing is clearer and easier to follow, addressing common feedback regarding the verbosity or robotic tone of Opus 5. The model is designed to prioritize important information upfront, fostering better collaboration over long sessions.

One early tester remarked that the model "writes the way I do," highlighting a shift toward a more intuitive user experience. This improved clarity serves a dual purpose: it enhances productivity for developers and researchers, and it acts as a safety benefit by making the model’s reasoning easier to audit and verify.

From a practical standpoint, the update also brings tangible service improvements. Anthropic has increased usage limits on Pro, Max, Team, and seat-based Enterprise plans to five hours. Furthermore, subscription users are receiving a rate limit reset that can be saved and utilized at their discretion, providing greater flexibility for bursty workloads.

Market Context and Competitors

The release of Opus 5.5 places Anthropic in direct competition with other leading models such as OpenAI’s GPT-6 Astra and GPT-5.6 Sol. While benchmarks show Opus 5.5 leading in specific categories like agentic coding and multidisciplinary reasoning, Anthropic acknowledges that margin differences at this level of capability are becoming less reliable indicators of real-world performance.

Internal testing suggests the gap between Opus 5.5 and Claude Fable 5.1 is narrower than benchmark scores imply. In areas such as business workflows, Opus 5.5 scored 40.0% on AutomationBench, slightly trailing GPT-6 Astra’s 41.4%. Similarly, in agentic scientific research, GPT-6 Astra led Terminal-Bench-Science 0.1 with a 64.6% score, compared to Opus 5.5’s 58.7%.

Despite these nuances, Opus 5.5 establishes itself as a cost-effective alternative for high-complexity tasks. Its ability to handle intricate coding projects and maintain high safety standards positions it as a strong contender in the enterprise AI market.

What Comes Next

Claude Opus 5.5 is the first installment in the Claude 5.5 family. Anthropic has confirmed that Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks. These subsequent releases are expected to bring similar improvements in performance, efficiency, and safety, making the advanced capabilities of the 5.5 family accessible to a broader range of users and applications.

Full details regarding the evaluation methodology and safety protocols are available in the official Opus 5.5 System Card.