Anthropic unveils Claude Opus 5.5 with 40% lower costs and top safety ratings

Key improvements include faster task completion—early adopters report migrating 680,000 lines of code in under a day (a task that would take weeks for a team) and auditing 200,000 lines of code in under three hours. Cost savings stem from a 20% reduction in token pricing ($4 for input, $20 for output) and a 60% cheaper cache-read mechanism, alongside a 51% lower cost for tasks like rewriting software (e.g., 9.5 hours vs. 12 hours for Fable 5.1). Benchmarks show Opus 5.5 outperforming OpenAI’s GPT-6 Astra in cost-efficiency by one-fifth for standard tasks, though GPT-6 Astra leads in scientific research agent work. Safety evaluations highlight a 85% drop in boundary-bypass attempts compared to Opus 5, with stricter restrictions for high-capability tasks like biology and cybersecurity—requiring approval for access.

The model is available across platforms (AWS, Google Cloud, Microsoft Azure) with increased usage limits for Pro, Max, Team, and Enterprise plans. Sonnet 5.5 and Haiku 5.5 are expected to follow within weeks. Anthropic notes benchmarks may understate Opus 5.5’s true capability due to safety measures rerouting high-risk tasks to older models.