Anthropic launches Claude Opus 5.5, first Claude 5.5 model, at 20% lower price and 40% lower run cost

Claude Opus 5.5 is positioned as Anthropic's most capable model for agentic coding, knowledge work, and long-running tasks, but the headline story is economics: at $4/$20 per million tokens it undercuts Opus 5 by 20% on sticker price, and Anthropic claims roughly 40% lower cost on typical workloads because the model generates faster and uses fewer tokens per task. The company says it matches Claude Fable 5.1 on most everyday work while extending usage limits for paying subscribers.

Mechanically, the release leans on token efficiency rather than raw scale, and adds a new safeguard stack: watermarking measures aimed at EU AI Act compliance, zero data retention options, and pre-release testing by external evaluators. Anthropic also partnered with Accenture on embedded evaluators for red-teaming and safety, signaling that safety tooling is being productized alongside the model.

Competitively, the timing is pointed. Opus 5.5 lands the same week OpenAI cut GPT-6 API prices by 50%+, making cheaper frontier intelligence the defining theme of the week. It is also Anthropic's first launch since Amodei's September 12 call for the industry to slow down — a tension critics immediately flagged, since a cheaper, faster model arguably accelerates adoption rather than pacing it.

What to watch: whether the 40% run-cost claim holds on real agentic workloads (where token consumption can balloon), and how the Fable 5.1-level performance framing lands with developers who expected a capability jump rather than an efficiency release. On Hacker News the launch dominated at 1,235 points, with a companion price/performance breakdown drawing its own debate over whether efficiency, not raw capability, is now the frontier battleground.