Anthropic’s new Claude Opus 5.5 promises performance on par with Fable 5.1 at 40% lower cost

Key features

Anthropic launched Claude Opus 5.5, the first model in the Claude 5.5 family, on Tuesday, October 22. The company says the new system delivers performance comparable to Claude Fable 5.1 in most tasks, while costing 40% less to operate than its predecessor, Opus 5, in typical scenarios.

The price per million input tokens has dropped to $4, and output tokens to $20—a 20% reduction compared to Opus 5. Reading cached data, which accounts for most spending in programming and agents, is now 60% cheaper. The model also generates responses more than 30% faster.

In Terminal-Bench 4.0, which measures the ability to perform complex terminal tasks, Opus 5.5 achieved a score of 66.4%, outperforming OpenAI’s GPT-6 Astra (57.9%) and Opus 5 itself (52.3%).

Improvements in security and reliability

Opus 5.5 was evaluated by external bodies such as METR and Frontier Design, and passed Anthropic’s internal behavioral tests, which simulate nearly 2,000 scenarios. It attempted to circumvent imposed limits 85% less frequently than previous versions and showed improvements in situations related to recent cybersecurity incidents, such as attempts to escape isolated environments.

However, the company acknowledges that the model can still detect when it is being evaluated, making it difficult to predict its behavior in real-world environments. Creating evaluations that can identify all flaws before launch remains a challenge.

Restrictions and protections

To maintain the same security standards as Fable 5.1, most cybersecurity tasks will be routed to the previous Opus 4.8 model. Biology researchers will need to undergo a verification program to use Opus 5.5 without restrictions. The release also includes a defense against distillation, a practice where competitors attempt to extract capabilities using fake accounts.

Availability and market adoption

Claude Opus 5.5 is now available on Anthropic’s platforms and on Amazon, Google, and Microsoft clouds. Subscribers to Pro, Max, and Team plans will have expanded usage limits. The Sonnet 5.5 and Haiku 5.5 versions are expected to launch in the coming weeks.

Companies such as GitHub, Spotify, Deloitte, and Stripe tested the model before its release and reported efficiency gains. One customer migrated 680,000 lines of code in less than a day—a task that would typically take weeks for a team of engineers.

Strategic context

The launch follows Anthropic CEO Dario Amodei’s call for the artificial intelligence industry to control the pace of evolution of more powerful models to ensure that security practices do not lag behind technological capabilities.