Claude Opus 5.5

Claude Opus 5.5, the new top model in the Claude 5.5 family, offers significant advancements compared to its predecessor. Early testers observed substantial performance improvements on complex tasks, such as migrating 680,000 lines of code in under a day. It excels at optimizing software by identifying and fixing inefficiencies, cutting web app load times in 39 out of 40 trials. Additionally, Opus 5.5 demonstrated superior graphics and polish when building a game from a single prompt.

The model also excels in safety, scoring the highest on automated behavioral audits among all models. It is less likely to perform irreversible actions or behave outside of its boundaries, while being more resistant to prompt injection attacks. Opus 5.5 will be deployed with similar safeguards as Claude Mythos 5.1 in biology and cybersecurity.

The model is more cost-effective and faster than its predecessor, with pricing at 40% less and 20% lower input and output token costs. Opus 5.5 generates output 30% quicker and also comes with increased five-hour usage limits on Pro, Max, Team, and Enterprise plans. Early testers found its communication style clearer and more natural, making it easier to follow over long sessions.

Claude Opus 5.5 leads in agentic coding, computer use, and knowledge work on our benchmarks. However, the gap between Opus 5.5 and Claude Fable 5.1 is narrower in real-world applications. In specific use cases, Opus 5.5 outperforms its competitors, such as auditing a 200,000-line codebase in under three hours, costing 51% less than its counterpart.

At default effort levels, Opus 5.5 matches GPT-6 Astra and beats GPT-5.6 Sol at a lower cost. It also has enhanced defenses against prompt injection attacks and includes a classifier, open-source sandbox, and code review to ensure safe operation.