Anthropic released Claude Opus 5.5 on Tuesday priced at $4 per million input tokens and $20 per million output, and says its safety classifiers can quietly reroute a request to Opus 4.8 or Opus 5 mid-workflow

A coding agent can go through a long run and get its answers from three different models without the developer knowing. Anthropic released Claude Opus 5.5 on 22 September 2026, and says the same safety classifiers already running on Claude Fable 5.1 will reroute Opus 5.5 requests to older Anthropic models when a prompt trips a cybersecurity, biology or frontier LLM flag. Most flagged cybersecurity requests go to Opus 4.8. Biology and frontier LLM flags go to Opus 5.

What the release actually costs

The New Stack reports Opus 5.5 at $4 per million input tokens and $20 per million output tokens, a 20 percent cut from Opus 5, with cache reads at $0.20 per million (down from $0.50) and cache writes at $5 per million (down from $6.25). Anthropic puts overall savings closer to 40 percent because Opus 5.5 uses fewer tokens per task and produces output over 30 percent faster than Opus 5. A separate fast mode in Claude Code and the Claude Platform runs up to 2.5 times faster at $8 per million input tokens and $40 per million output tokens.

The scores it is claiming

On Anthropic's own numbers, Opus 5.5 scores 66.4 percent on Terminal-Bench 4.0 against Fable 5.1's 55.8 percent, and 54.4 percent on FrontierCode v1.1 against 50.3. On CursorBench 4.0 it reaches 57.8 percent, ahead of Fable 5.1 at 51.8 and GPT-5.6 Sol at 41.7. Anthropic says a few points at this level do not translate into a noticeable difference in real work.

Early enterprise numbers are the more useful comparison. The New Stack reports that Box measured Opus 5.5 at about a third of the tokens Opus 5 used on the same evaluations, with answers 40 percent less verbose. GitHub tested it inside Copilot CLI and VS Code and reports Opus 5.5 finished more terminal tasks than Opus 5 in less than half the steps. Deloitte reports Opus 5.5 at its lowest-effort setting caught 72 percent of known bugs in code reviews against 56 percent for Opus 5 at high effort.

The silent reroute is the detail to design around

For any team building an agent workflow, the safety-classifier reroute is the part that needs an architectural decision. A request sent to Opus 5.5 can be handled by Opus 4.8 or Opus 5 instead when a classifier fires, so two turns in the same run can come back from two different models. Evals built on the assumption of a single model will not see this variance, and downstream steps that rely on Opus 5.5 capabilities may run against the older one without the caller being told which was used.

Vetted organisations can apply to Anthropic's Life Sciences Verification Program to use Opus 5.5 without the biology classifier, and the company says the Cyber Verification Program will expand to include Opus 5.5 in the coming weeks with three tiers of trusted access. Frontier Design and METR evaluated the model before release. Opus 5.5 is the first model in the Claude 5.5 family, with Sonnet 5.5 and Haiku 5.5 to follow.

Source

Anthropic, Introducing Claude Opus 5.5, 22 September 2026. Reporting, pricing detail and enterprise numbers from Amanda Caswell, The New Stack, Anthropic releases Opus 5.5 and cuts pricing by 20 percent. Your agent calls might secretly get routed to an older model., 22 September 2026.