Anthropic released Claude Opus 5.5 on September 22, 2026, introducing a model that significantly alters the “Claudish” writing style while reaching performance parity with the high-end Fable 5.1 tier. The model features input tokens priced at $4 per million and output tokens at $20 per million.
Data from Arena suggests a major shift in how the model structures language. Specifically, Claude Opus 5.5 has reportedly reduced its use of em dashes by 95%, dropping from 15.2 to 0.8 per 1,000 words. Semicolon usage also fell sharply, moving from 6.10 to 1.64 per 1,000 words. These changes appear to be part of a broader effort to de-jargonize AI output and move away from the overly punctuated style often associated with earlier large language models.
Despite the cleaner punctuation, the model has become more verbose in its total output. While the average sentence length in Opus 5.5 fell by 17%—moving from 12.14 to 10.03 words—the average overall answer length increased from 453 to 481 words. This 6% increase in total word count suggests that while the model is using simpler, punchier sentences, it is providing more detailed explanations or including more steps in its reasoning process.
Performance and Cost Efficiency
Claude Opus 5.5 is designed to match the capabilities of the previously “frontier” Fable 5.1 model. According to Anthropic, the new model is approximately 40% less expensive to operate than its predecessor on typical workloads. This efficiency is partly driven by a massive reduction in cached token costs; cache reads now cost $0.20 per million tokens, representing a 60% decrease.
The model maintains a 1 million token context window and features a knowledge cutoff of June 2026. It also marks a change in how users are expected to interact with the system. Because the model now incorporates native “thinking” processes as a permanent feature—a move reportedly intended as an anti-distillation measure—Anthropic has removed “think carefully” from its recommended prompting playbook. The model is essentially designed to perform that reasoning by default.
Strategic Alignment
The release follows a September 12, 2026, essay by Anthropic CEO Dario Amodei titled “We Must Pace the Frontier,” which outlined the company’s approach to scaling high-performance models while maintaining safety and usability standards. Opus 5.5 serves as the flagship for this strategy, balancing high-end reasoning with a more human-like writing tone that avoids “AI slop” patterns.
The current pricing structure and performance metrics position Opus 5.5 for high-volume professional use cases where cache efficiency and natural language quality are primary concerns.
