Anthropic has unveiled Claude Opus 5.5, marking the first release in the Claude 5.5 series. The company describes this launch as a flagship upgrade focused on programming, agent-based scenarios, analytics, and security.

Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family.

It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5. pic.twitter.com/Q9C2VKQ79f

— Claude (@claudeai) September 22, 2026
 

The company claims that this model operates at a level comparable to Claude Fable 5.1 for most tasks while consuming 40% less computational resources than Opus 5.

According to the developers, the model has significantly improved in handling complex tasks: it excels in migrating large codebases, conducting code audits, optimizing applications, and preparing analytical reports. In one internal test, Opus 5.5 completed the migration of 680,000 lines of code in under a day—a task that Anthropic asserts would have taken weeks for an engineering team.

New Claude 5.5 Series

Opus 5.5 is the inaugural release in the new Claude 5.5 series. In the coming weeks, Anthropic plans to launch Sonnet 5.5 and Haiku 5.5, which are expected to feature similar enhancements in performance, efficiency, and security.

The company emphasizes that the model was developed following CEO Dario Amodei's call for advanced AI systems to evolve alongside measures for control and risk assessment.

Cost-Effective and Efficient

Anthropic asserts that Opus 5.5 is more economical not only due to a lower token cost but also because it uses fewer tokens per task.

Source: Anthropic.

According to the company’s calculations, this leads to approximately 40% savings on typical workloads. Furthermore, the model reportedly generates responses over 30% faster than Opus 5.

For users on Pro, Max, and Team subscription plans, Anthropic has increased usage limits by five hours and introduced the option to save and later utilize the reset limit.

Focus on Coding and Agent-Based Scenarios

Anthropic highlights that Opus 5.5 is particularly adept at agent-based coding and tasks requiring long-term autonomous operation. In one test, the AI successfully optimized a web application, reducing page load times in 39 out of 40 cases. In another test, it outperformed other models in graphics quality and "polishing" of a game created from a single prompt.

The company places special emphasis on corporate applications, stating that Opus 5.5 is well-suited for financial analysis, business reporting, and data research. Anthropic also notes that the AI shows strong results in benchmarks for intellectual tasks and agent coding, although they acknowledge that high-level testing may not accurately reflect real-world differences between models.

Source: Anthropic.

Enhanced Security Measures

With the increased capabilities of Opus 5.5, Anthropic has also strengthened its restriction systems. The company reported that Opus 5.5 achieved the highest scores among all tested models in an automated "behavioral audit"—an internal assessment of consistency and safe behavior across thousands of simulated scenarios.

Anthropic states that the model:

  • is less likely to take irreversible actions;
  • is better protected against prompt injection;
  • less frequently exceeds its prescribed limits;
  • shows improvements in honesty and adherence to instructions.

However, the company admits that there are still no perfect evaluation methods, as models can "understand" that they are being tested, making real-world scenarios more complex than laboratory conditions.

Controlled Cyber and Bio Scenarios

Due to its high efficacy in cybersecurity and biology, Anthropic has implemented protective restrictions on Opus 5.5 comparable to those used for its most powerful models.

Specifically:

  • most cyber requests will be redirected to other models;
  • access to research tasks in biology will be granted only to verified organizations through the Life Sciences Verification Program;
  • the Cyber Verification Program will expand in the coming weeks for verified cybersecurity professionals.

Additionally, Opus 5.5 has received "saved reasoning" protection, which limits attempts to extract the model's internal reasoning through the API.

Notably, in September, Anthropic announced a partnership with consulting firm Accenture to conduct independent assessments of its AI models.

Follow ForkLog on social media

Telegram (main channel) Facebook X