Anthropic · updated 2026-08-29
Run Claude Opus 5: providers, pricing and flat-rate access
Anthropic's high-end Claude 5 tier below Fable — top-shelf reasoning for hard agentic work, priced at $5/$25 per M tokens (a Fast variant runs at double rates for lower latency).
Opus 5 without an Anthropic plan means the Anthropic API per-token ($5/$25 per M tokens — premium rates that compound hard in agent loops) or a flat-rate Messages-API provider: Standard Compute includes Opus 5 (and Opus 5 Fast) in its routed pool at a fixed monthly price, reachable from Claude Code via a base-URL swap.
Your options, honestly
Claude Pro / Max$20–200/moFirst-party; Opus access sits on the upper tiers, still metered by session windows and weekly caps.
Anthropic APIPay-per-token$5/$25 per M tokens; a heavy Opus-only agent multiplies a premium rate by growing context — this is where per-token pain is sharpest.
Standard Compute$39+/mo flatOpus 5 and Opus 5 Fast in the routed pool for hard reasoning steps — the frontier step without the frontier bill. Routing decides when Opus-class fires; no pinning.
Bedrock / VertexPay-per-tokenEnterprise cloud billing, same economics; useful when procurement requires it.
the flat-rate trade-off
Opus is exactly the model class flat-rate routing exists for: too expensive to run on every step, too good to skip on the hard ones. Routing reserves Opus-class for the steps that need it — which is also the honest caveat: you can't pin Opus-only, and flat price is not infinite compute.
Running it in your agent
Claude Code: ANTHROPIC_BASE_URL=https://api.stdcmpt.com puts Opus-class in reach without windows. OpenAI-compatible agents get it through the routed 'standardcompute' model on hard steps.
FAQ
What's Claude Opus 5 Fast?
A lower-latency serving tier of Opus 5 at double the per-token rate ($10/$50 per M tokens) — for when response time matters more than cost. In a routed flat-rate pool, latency-sensitive hard steps can land on it without a separate bill.
Can I run Opus 5 on a flat monthly price?
Anthropic's own flat plans are windowed app/CLI tiers. For API-style agent use, the practical flat-rate route is a multi-model plan that includes Opus 5 in routing — fixed price, routing decides when it's used.
More: how flat-rate LLM APIs work · per-agent guides for OpenCode, Claude Code and Cline · model comparisons