Anthropic released Claude Opus 5.5 today, September 22, 2026. PressBot 1.9.0 ships the same day. claude-opus-5-5 is already selectable in two places: the Pro admin agent and PressBot Shield.
It is not in the visitor chatbot dropdown. That was deliberate.
A cheaper Opus is still an Opus
Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens. Opus 5 still sits at $5 / $25 and remains labeled previous. Anyone already paying Opus 5 rates will notice the cut.
It is still flagship-tier money. Your public widget answers the same shipping question all week for visitors who never log in. A $20-per-million output rate does not belong behind “do you ship to Canada?”
The visitor chatbot default stays Claude Haiku 4.5. We did not add Opus 5.5 to that dropdown, and we do not plan to.
Thinking tokens land on the output line
Adaptive thinking is always on in Opus 5.5. You cannot switch it off. Anthropic’s default effort is medium, and PressBot does not send a custom effort parameter, so the model decides how much to reason. Anthropic bills that reasoning as output.
At $20 per million output tokens, 10,000 tokens of reasoning costs $0.20 before a single visible word. Fine on one hard admin task. Across thousands of anonymous visitor sessions you do not control, it is a line item on your Anthropic invoice.
Behind the admin login, one operator sends requests they chose to send. Variable reasoning cost is acceptable there. On the public widget, it is not.
Where Opus 5.5 earns the rate
An agency operator comes back after a quiet weekend. Several things changed on a client site. They want the relevant information gathered, connected, and turned into a recommendation.
That is the work where a shallow answer is worse than no answer. The value is reasoning across several inputs to one conclusion, with an operator who reads it before acting. That is the job we put Opus 5.5 on the agent for.
Shield follows the same split. It sits on the operator side. When the question is “what does this mean for my site,” a careful read beats a fast one. Opus 5.5 is selectable there too.
Another use case is using Opus 5.5 as the Writer or Critic Agent Model for the Blog Post creation content pipeline. It follows directions better, and also has improved writing capabilities so it’s a natural fit if you want to generate superior Blog posts.
Your saved pins stay put
The admin agent default stays Claude Sonnet 5. If you pinned Opus 5, it stays pinned. PressBot 1.9.0 does not auto-migrate saved Opus 5 selections to 5.5.
Opus 5 remains selectable at $5 / $25 with a previous label, so you see the choice instead of waking up to a different model. Swapping models changes behavior, latency, and spend. You should make that change on purpose.
How we’d pick for a real session
Anthropic positions Opus 5.5 as the everyday start-here model. Fable 5.1 remains the flagship for jobs where Opus 5.5 still falls short. PressBot keeps Fable 5.1 agent-only, for the same reason Opus 5.5 is agent-only.
Leave Sonnet 5 on the agent for routine admin sessions. Most sessions do not need more.
Move to Opus 5.5 when a wrong call costs money or a client relationship: security questions, multi-step restructures, anything that touches a live client site. If you already pay Opus 5 rates, run one real workload against 5.5 this week.
Keep Opus 5 only if you measured it on your own workflows and do not want to re-test yet. “It’s what we had” is not a reason.
Reach for Fable 5.1 after Opus 5.5 has failed at the task, not before. For the visitor widget, stay on Haiku 4.5.
Also in 1.9.0
An optional Modern visitor layout with Home and Messages views and a public-only video slot. Existing sites stay on Classic until an admin opts in. Preview it on the PressBot tour.
Run one task on 5.5 today
PressBot installs from the Freemius zip at pressbot.io. It is not listed on WordPress.org. Pro plans are still $29, $59, and $99 per year. Opus 5.5 runs on your own Anthropic key, so model spend goes to your Anthropic account, not to us.
If you already run Pro, update to 1.9.0. Open the admin agent’s model selector, pick claude-opus-5-5 for one real task you’d normally send to Opus 5, then compare the output and your Anthropic usage page. Watch the output token count — that is where the thinking lands. Full notes are in the changelog.
Evaluating Pro for the first time? Start at pressbot.io/pro.
