Claude Opus 5.5 in Claude Code: Claude AI API pricing, changes and effort

Claude Opus 5.5 was released on 22 September 2026. Starting with Claude Code 2.1.280, the opus alias points to it when connecting through the Anthropic API, Claude Platform on AWS, Amazon Bedrock and Google Cloud; in Microsoft Foundry, it points to Opus 4.6. The model your session uses also depends on the settings and on whether it is a new or resumed session, so it is worth checking the model. Opus 5.5's default reasoning level, or effort, is medium rather than high, which is the default for most models that support effort. The previous global level from user settings does not apply to it.

The facts below come from the official Claude Code documentation on model configuration and the Opus 5.5 and Opus 5 model pages on Anthropic's website. All three documents were checked again on 23 September 2026.

Ailoot documentation: API, keys and connection setup.

What changed in Claude Code with the release of Opus 5.5#

Opus 5.5 requires Claude Code version 2.1.280 or later. The documentation recommends updating with claude update. How to check the version and update the application is explained in the guide on updating Claude Code to the latest version.

Selecting opus. This is a generic model name (an alias) without a version number. It points to the recommended version and changes over time. Since version 2.1.280, opus points to Opus 5.5 when connecting through the Anthropic API, Claude Platform on AWS, Amazon Bedrock and Google Cloud; it previously pointed to Opus 5. The exception is Microsoft Foundry, where opus points to Opus 4.6.

The Default option. This is the model Claude Code uses if you have not selected a model yourself. On Pro, Max, Team and Enterprise plans, and when using the Anthropic API, this is now Opus 5.5. Before 2.1.280, the Default option was Sonnet 5 on Pro and Team Standard, and Opus 5 on Max, Team Premium and Enterprise. If you are on Pro and have not set a model anywhere, Default in a new session after the update will be Opus 5.5 instead of Sonnet 5. If an organization administrator on an Enterprise plan has set a default model, Default points to that model.

A pinned version. If a full name is specified in the settings or at startup, such as claude-opus-5 for Opus 5 or claude-opus-5-5 for the new model, updating does not switch the session to another version. You can also pin a version with the ANTHROPIC_DEFAULT_OPUS_MODEL variable.

Claude AI API pricing: Opus 5.5 versus Opus 5, caching and context#

This comparison is based on Anthropic's official model pages. Prices are the Claude API base rates per million tokens and apply to use through an API key. These pages say nothing about subscription pricing.

Caching. A cache read usually costs 10% of the base input price; for Opus 5.5, it costs 5%. Compared with Opus 5, the cache-read price has fallen by 60%, from $0.50 to $0.20. The input-token price has fallen by 20%, from $5 to $4.

Context. 1M tokens, the same as Opus 5. Anthropic estimates this at approximately 555 thousand words with the current tokenizer. When connecting directly to Anthropic, Opus 4.7 and later models, including Opus 5.5, use a 1M window on any plan, including Pro. You do not need to select a separate [1m] option for this.

Opus 5's Legacy status means the model remains available, but Anthropic recommends moving to Opus 5.5. Opus 5 was released on 24 July 2026; it is covered in the guide on whether to switch to Claude Opus 5.

What effort level Opus 5.5 starts with in Claude Code#

Effort controls the depth of reasoning: the model decides for itself whether to think at each step and for how long. A low level is faster and cheaper for simple tasks; a high level provides deeper reasoning for complex ones. You cannot disable reasoning on Opus 5.5: the session toggle and the corresponding settings have no effect on it.

Alongside the levels, the /effort menu has an ultracode option. This is not a model level, but a separate Claude Code setting: it sends xhigh to the model and also starts dynamic workflows for large tasks. If ultracode is disabled, Claude Code selects the session's effort level in the following order, taking the first applicable value:

1. An explicit choice: the CLAUDE_CODE_EFFORT_LEVEL environment variable, starting with the --effort flag, or the /effort command in the session.
2. Settings: the level saved for this model, or the effortLevel key.
3. The model's default level: high for all models that support effort except Opus 5.5 (medium) and Opus 4.7 (xhigh). If the organization has set an effort level for its default model, that level applies when using that model.

What does not carry over. Previously, the /effort command wrote one global level to the effortLevel key in the user settings file. Claude Code now stores the level separately for each model. The old global key still applies to Opus 5, Fable 5.1 and earlier models, but not to Opus 5.5. If none of the other sources listed above sets the level, Opus 5.5 starts with medium until you choose a level for it through /effort or in /model. The exception is effortLevel set in project, local or managed settings, or supplied through --settings: it applies to all models, including Opus 5.5.

What medium means. In Claude Code's description of the levels, medium reduces token use for work where cost matters and some intelligence can be sacrificed. High is described as a balance between token use and intelligence. According to the documentation, the default level suits most coding tasks; changing it is recommended when you need a different balance. Whether it suits your tasks needs to be tested on your own work.

How to check the model and effort in your session#

1. Update Claude Code to 2.1.280 or later. Without this, Opus 5.5 is unavailable in Claude Code.
2. Look at the session header. Next to the model name, Claude Code shows the current effort level, for example, “with low effort”. At startup and when you change the level, it also appears briefly in the bottom line.
3. Open /model. With no argument, the command shows a list of models. For models that support effort, the list has a slider, which you move with the left and right arrow keys. Enter switches the model and saves it as the default for new sessions; the s key switches it only for the current session.
4. Change the level through /effort if needed. With no arguments, the command opens the slider; with a level name, such as /effort high, it sets that level immediately. Enter or a level typed after the command saves the choice for future sessions of that particular model; the s key changes it only for the current session. /effort auto resets the saved level for the active model.

For example, to save high for future sessions with Opus 5.5, run the command below in a session using that model. The CLAUDE_CODE_EFFORT_LEVEL variable or the --effort flag at startup will still take priority over the saved level.

/effort high

If the model is not the one you selected

Higher-priority settings can override a choice made through /model:

• model in project or managed settings. This value is reapplied at every startup, and the startup header shows which file set it.
• The ANTHROPIC_MODEL environment variable in your shell.
• The default model set by an organization with permission to override the user's choice.

A resumed session (claude --resume, --continue or /resume) continues using the model saved in that session. If you want to try Opus 5.5, start a new session or switch the model through /model.

If you previously selected a model using the guide on Sonnet or Opus in Claude Code, bear in mind that it was written before Opus 5.5 was released. The previous Opus version is covered in the guide on Opus 4.8 in Claude Code.

Brief recap#

Claude Opus 5.5 was released on 22 September 2026. Starting with Claude Code 2.1.280, the opus alias points to it in the Anthropic API, Claude Platform on AWS, Amazon Bedrock and Google Cloud; in Microsoft Foundry, it points to Opus 4.6. The Default option selects Opus 5.5 on Pro, Max, Team, Enterprise and the Anthropic API unless the model is overridden in the settings. On Pro and Team Standard, it previously selected Sonnet 5. At Claude API rates, the new model is cheaper than Opus 5: $4 and $20 per million input and output tokens, compared with $5 and $25, and cache reads cost $0.20 rather than $0.50; the context window remains 1M. Opus 5.5's default effort level is medium, and the old global effortLevel from user settings does not apply to it. Check the model and level in the session header and choose a level through /effort if needed.

If you need access to AI models through Ailoot, explore Ailoot.