Anthropic's Claude Opus 5.5 optimizes for extended coding sessions with 40% lower costs and enhanced context handling.
Anthropic has launched Claude Opus 5.5, a new iteration of its Claude AI models designed to optimize for long-running, context-heavy coding sessions. According to the company, Opus 5.5 reduces costs by approximately 40% compared to its predecessor, Claude Opus 5, making it a competitive option for developers handling complex, open-ended tasks.
With software development increasingly leaning on AI for both day-to-day operations and high-level problem-solving, the ability to process extended context efficiently has become critical. Anthropic’s aggregate data from March to September 2026 shows that developers are asking their models to do more: context per request has grown 2.6x, and input-to-output token ratios have jumped from 189:1 to 324:1. Opus 5.5 is built to handle this shift, particularly for scenarios involving agentic coding, tool integrations, and fewer user interruptions.
One of the key cost-saving mechanisms is a significant reduction in the price of cached token reads—down 60% compared to Opus 5. Cached tokens account for a substantial portion of costs in agentic coding workflows, where reusing previously computed context can lead to significant efficiency gains. Anthropic claims Opus 5.5 can read cached tokens at just 20% of the cost of competing models, providing a strong economic incentive for developers to adopt the new model.
5 also introduces enhanced cache management capabilities. For example, developers can now set a one-hour cache lifetime on API keys and cloud providers, while forked subagents inherit parent cache context, reducing redundant token usage.
Source link
