Anthropic announced Claude Opus 5.5 on September 22, 2026, as the first release in the Claude 5.5 family. The company says its listed prices are $4 per million input tokens, $20 per million output tokens, and $0.20 per million cache-read tokens; it characterizes the result as a 40% cost reduction on typical workloads versus Opus 5. It also says output generation is more than 30% faster than Opus 5 and that five-hour usage limits are increasing on Pro, Max, Team, and seat-based Enterprise plans.
The mechanism is straightforward: agentic coding and other long-running workflows repeatedly consume context and cached state, so cheaper cache reads, lower token prices, and faster generation reduce both direct spend and time-to-completion. Anthropic pairs that efficiency claim with safety positioning: it says Opus 5.5 is less likely to take hard-to-reverse actions or exceed assigned boundaries, and more resistant to prompt injection than Opus 5. This matters now because these changes alter the cost and reliability assumptions for teams deciding whether to put a frontier model into recurring production workflows.
The first implication is that model selection for agents may increasingly hinge on the combined cost of tokens, cache use, latency, and operational guardrails rather than a single benchmark rank. A second implication is that higher subscription limits and a saveable rate-limit reset can change how individual users schedule intensive work. The evidence boundary is that the performance, efficiency, and safety claims are Anthropic’s own public assertions, including its chosen evaluations and test settings; they are not an independent market comparison. The first watch condition is whether independent long-horizon agent evaluations reproduce the claimed cost, speed, and boundary-following advantages in production-like environments. Watch also for the announced Sonnet 5.5 and Haiku 5.5 releases, plus the promised expansion of verified cybersecurity access.