It Is No Longer Just a Model
Anthropic shipped Opus 5 on July 24. Within two weeks it had crawled into GitHub Copilot and Amazon Bedrock.
The developer's billing line changed quietly.
The Price, On the Table
- API per million tokens: $5 input, $25 output — same as Opus 4.8
- Fast mode: about 2.5x speed, but double the base price
- 1M-token context, adjustable effort, default strongest model on Max tier
How It Spread
Into Copilot
GitHub added Opus 5 to the model picker across Pro+, Max, Business, Enterprise. Target workload: complex, long-running, tool-calling coding.
But there is a catch: picking Opus 5 in Copilot is billed by usage (provider API list price), not bundled in the seat fee. Business and Enterprise admins must enable the policy before users can select it.
Into Bedrock
On AWS the model ID is anthropic.claude-opus-5, with 1M context, 128K max output, adaptive thinking, computer use, prompt caching.
But Bedrock Runtime does not support Responses or Chat Completions — only Invoke, Converse, Messages, Text. Check the interface before migrating; do not assume.
Three Traps in Practice
Trap one: the bill. Enabling Opus 5 in Copilot is usage-billed, outside the seat fee. If the team does not track it separately, finance will find you at month-end.
Trap two: the safety classifier. Opus 5 added stronger guardrails on high-harm cyber content; some security-adjacent requests get blocked. Teams building security tooling must test this before production.
Trap three: automatic fallback. Requests flagged by the safety classifier no longer hard-fail — they route to another model. Claude.ai/Code/CoWork default to Opus 4.8 as fallback. That means a request may "complete on a different model," so your production logs must record the actual model chosen, or debugging goes blind.
Who Should Use It, Who Should Save
- Complex long-horizon coding, agent workflows, multi-round verification → Opus 5 is worth it, but test on a small repo and track cost separately
- Daily completion, simple rewrites, high-frequency short tasks → use Sonnet or a faster Flash tier; do not waste the $25 output price
- Already on the AWS stack → integrate via Bedrock, mind the interface gap and region/account access
Why This Matters to You
You are a tech lead, solo dev, or small-team boss — picking models is not "newest wins."
Opus 5 is strong but pricey. Put it on "expensive long tasks," put cheap models on "high-frequency short tasks." Mixing is the answer. Blanket Opus 5 across everything will terrify your bill.
Also, automatic fallback is a good design, but it forces one habit: log which model actually ran. Build it now.
Bottom Line
Opus 5 matters not just because it is "a bit stronger." It is the first top-tier model that truly became a deployment option — into the editor, into the cloud, with auto-fallback. The developer's job shifted from "pick a model" to "place models in the right slots and account for the cost."