A New Entry in the Claude 5.5 Family
Anthropic has released Claude Opus 5.5, described as the first model in its new Claude 5.5 family and made available on the day of announcement. According to Anthropic, the model performs at the level of Claude Fable 5.1 for most tasks while costing around 40% less to run than Opus 5. That combination of near-top-tier quality at a lower running cost is the headline reason developers and businesses are paying attention.
The model is built specifically for long-running agentic coding and knowledge work. In plain terms, it is aimed at tasks where an AI has to keep working across many steps, call tools, read large volumes of material, and stay coherent over an extended session rather than answering a single quick question.
Pricing and Core Specifications
Claude Opus 5.5 is priced at $4 per million input tokens and $20 per million output tokens. For context on how it sits within the family, the docs list Claude Fable 5.1 at $10/$50, Claude Sonnet 5 at $2/$10, and Claude Haiku 4.5 at $1/$5. The Batch API offers a 50% discount on both input and output.
- Context window: 1M tokens
- Max output: 128K tokens (up to 300K on the Batch API with a beta header)
- Thinking: Adaptive and always on, with a default effort of medium
- Knowledge cutoff: June 2026
- Released: September 22, 2026, with retirement not sooner than September 22, 2027
It is available across the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and the Claude Platform on AWS, all under the model ID claude-opus-5-5.
Breaking Changes Developers Should Know
If you already have code running on Claude Opus 5, four changes may affect you. Thinking can no longer be disabled. Forced tool use now returns an error. Thinking blocks are tied to the specific model and conversation. And on the Claude API and Google Cloud, the earlier computer_20251124 computer use tool is not accepted. The first three of these also apply on Claude Fable 5.1.
There is one more subtle change that does not fail requests but alters the response shape. Text that appears between tool calls now comes back inside thinking blocks whose text is empty at the default display setting. If your application streams that text to users as progress updates, it will go quiet between tool calls until you set a display value that returns the text.
A Practical Example
Imagine a support engineering team that wants an assistant to review a large codebase and open a fix. A workflow using Opus 5.5 might look like this:
- Feed the assistant a repository documentation set and recent issue reports, staying within the 1M-token context window.
- Set the effort parameter to control thinking depth. For a straightforward bug, keep it lower to save cost and time; for a tricky architectural change, raise it.
- Let the model call tools to search files, run tests, and draft a patch across many steps.
- Because forced tool use now returns an error, update any older code that relied on it so the model chooses tools on its own.
- If your interface shows live progress, make sure you set the display value that returns inter-tool-call text, otherwise the status panel will appear frozen between actions.
This kind of setup shows why the model is positioned for agentic work: it is meant to keep going across long sessions rather than handling one prompt at a time.
Where It Fits
For teams that found Opus 5 capable but expensive, the lower running cost of Opus 5.5 makes extended agentic tasks more affordable. Faster, cheaper models like Sonnet 5 and Haiku 4.5 remain better suited to quick, high-volume requests, while Opus 5.5 targets deeper work that benefits from a large context and always-on reasoning.
Final Thoughts
Brain.mt can help you using AI for your business. Contact me for more information. I also offer dedicated workshops and training about this subject, whether you want to plan an agentic coding workflow, compare model costs, or update existing integrations to match new model behaviour.
