Anthropic Unveils Claude Opus 5.5: Enhanced Coding Capabilities at 40% Lower Cost

Key Highlights

• Claude Opus 5.5 arrives with performance nearly matching Anthropic’s premium models while reducing operational expenses by approximately 40% compared to Opus 5.
• Pricing is set at $4 for each million input tokens and $20 for each million output tokens, representing a 20% reduction from Opus 5’s rates.
• According to Anthropic, Opus 5.5 excels in specialized domains such as agentic coding, computer operation, and professional knowledge tasks.
• Independent assessments by Frontier Design and METR were conducted prior to public release, with enhanced protections for cybersecurity and biological applications.
• Access to Claude Opus 5.5 is now available via AWS, Google Cloud, Microsoft Azure, and Anthropic’s proprietary platform.

Anthropic has unveiled Claude Opus 5.5, marking the debut entry in its Claude 5.5 series, emphasizing enhanced coding abilities, professional applications, and reduced operational expenses. According to the company, this model delivers performance levels rivaling its premium offerings while cutting running costs by roughly 40% versus Opus 5.

This announcement follows recent statements from Anthropic CEO Dario Amodei urging the AI sector to moderate frontier model development, allowing safety mechanisms to align with emerging capabilities. Anthropic reports that Opus 5.5 completed third-party assessments prior to deployment and incorporates protective measures typically associated with its most advanced systems.

Reduced Pricing Structure for Claude Opus

The pricing for Claude Opus 5.5 stands at $4 per million input tokens and $20 per million output tokens. This represents a decrease from Opus 5’s rates of $5 and $25 respectively, while cache-read costs have dropped from $0.50 to $0.20 per million tokens.

According to Anthropic, the model requires fewer tokens for numerous operations, contributing to the overall 40% cost reduction for standard workloads. The company reports that output generation speeds have improved by over 30% compared to Opus 5.

The company has also expanded five-hour usage caps for Pro, Max, Team, and seat-based Enterprise subscribers. Anthropic announced that subscription members will gain access to a rate-limit reset feature that can be stored and utilized at a later time.

An accelerated mode has been introduced through Claude Code and the Claude Platform. This high-speed option operates at up to 2.5 times standard velocity, with premium pricing of $8 per million input tokens and $40 per million output tokens.

Focus on Software Development and Professional Applications

According to Anthropic, Claude Opus 5.5 demonstrates exceptional capability in extended software engineering projects including migrations, code reviews, and defect resolution. In preliminary testing, the model allegedly finished a 200,000-line codebase examination in less than three hours, contrasting with over 20 hours required by Opus 5.

Based on Anthropic’s disclosed Terminal-Bench 4.0 metrics, Opus 5.5 achieved 66.4%, outperforming GPT-6 Astra’s 57.9% and GPT-5.6 Sol’s 37.3%. The company emphasizes that benchmark outcomes vary based on model configuration and testing methodology, noting that practical performance gaps may be less pronounced than numerical scores indicate.

The company additionally emphasized advancements in professional knowledge work and enterprise applications. Using the GDPval-AA v2.1 benchmark, which assesses professional performance across 44 career fields, Opus 5.5 registered 1846 Elo, surpassing Anthropic’s earlier systems.

According to Anthropic, preliminary enterprise users including GitHub, Deloitte, Stripe, and additional organizations observed enhancements in software creation, analytical work, and written correspondence. These represent company-supplied assessments and client feedback rather than independently verified measurements across all practical scenarios.

Enhanced Security and Safety Protocols

Prior to launch, Claude Opus 5.5 underwent evaluation by independent organizations including METR and Frontier Design. Anthropic reports the model produced its most robust performance on the company’s internal automated behavioral assessment.

The company stated that Opus 5.5 demonstrated approximately 85% fewer attempts to circumvent containment protocols in specialized testing compared to Opus 5 or Claude Mythos 5.1. Anthropic further claims the model exhibits improved resistance to prompt-injection vulnerabilities relative to Opus 5.

Given its proficiency in cybersecurity and biological sciences, the model incorporates supplementary protective measures in these domains. Anthropic is extending verification initiatives that grant authorized researchers and cybersecurity specialists expanded access for sanctioned projects.

Claude Sonnet 5.5 and Claude Haiku 5.5 are anticipated to launch within the upcoming weeks. Opus 5.5 is currently accessible through Anthropic, Amazon Web Services, Google Cloud, and Microsoft Azure.