Claude Opus 5: Agentic Coding and Computer Use at Unchanged Pricing

What happened

Anthropic has launched Claude Opus 5, its new flagship model at the Opus tier, replacing Claude Opus 4.8. The release positions Opus 5 as approaching the intelligence of Anthropic's higher-tier Claude Fable 5 model whilst retaining the existing Opus pricing structure of $5 per million input tokens and $25 per million output tokens. Opus 5 is now the default model within its tier on Anthropic's platform.

Anthropic describes Opus 5 as purpose-built for frontier-class agentic coding and computer-use tasks — capabilities that allow the model to autonomously navigate software environments, execute multi-step workflows and interact with digital interfaces on a user's behalf.

Why it matters

Agentic AI — models that can plan, act and course-correct across extended tasks without constant human input — is rapidly becoming the infrastructure layer beneath customer-facing services. When an AI can operate software autonomously, it shifts from being a conversational assistant to a genuine service agent: booking, resolving, escalating and completing on behalf of customers. For CX leaders, this changes the design brief entirely. The question is no longer "what should the AI say?" but "what should the AI do, and on whose authority?"

From a behavioural-economics standpoint, agentic systems introduce new trust dynamics. Customers must decide how much autonomous action they are comfortable delegating — a calculus shaped by perceived competence, past experience and the reversibility of the actions taken. Service designers who ignore this psychological dimension risk deploying capable technology that customers nonetheless refuse to use.

By the numbers

• $5 per million input tokens — Opus 5 pricing, unchanged from its predecessor.
• $25 per million output tokens — output pricing, also held flat at the Opus tier.
• Half the price of Claude Fable 5, according to Anthropic's own positioning, for comparable intelligence benchmarks.

The Renascence take

Most commentary on Opus 5 will focus on benchmark scores and developer economics. What deserves equal attention is the service-design implication of "computer use" as a standard model capability — and how few organisations are structurally ready for it.

The real frontier here is not model intelligence — it is institutional readiness. Agentic AI acting inside live customer journeys demands governance frameworks, clear escalation paths and explicit customer consent architecture that most CX operations have not yet built. The behavioural principle to keep front of mind is autonomy calibration: customers will tolerate AI acting on their behalf only to the degree that they feel in control of the boundaries. Operators who lead with capability without leading with transparency will find adoption stalls not because the technology fails, but because trust does. The contrarian move right now is to slow down the deployment roadmap just long enough to design the consent and override experience properly — that investment will compound.