Claude Opus 5 to Opus 5.5: Model ID, Claude Code Setup, and API Migration
Start Here: Changing the Model ID Is Not the Whole Migration
The Claude API model ID for Opus 5.5 is claude-opus-5-5. When moving an Opus 5 integration, review four breaking changes: thinking cannot be disabled, forced tool use returns an error, thinking blocks are tied to the model and conversation, and an older computer-use tool is rejected on specific official platforms.
There is also a separate change that does not fail the request: text between tool calls comes back in thinking blocks whose text is empty at the default display setting. A progress interface that relies on that text can go quiet even while the request continues.
These are Anthropic's Opus 5.5 specifications. They are not a statement that llmapi.pro has tested every feature or exposes every official limit. For llmapi.pro, check the model's billing mode and your client's protocol separately before switching.
Check Access, Billing Mode, and Protocol First
The llmapi.pro subscription page lists both claude-opus-5-5 and claude-opus-5. It identifies GET /v1/models as the authoritative subscription-availability list. Use the exact request ID, not a display name.
The model-selection guide separates three settings that should not be conflated: model name, protocol compatibility, and billing mode. In particular:
- Monthly subscription: use a subscription key. Supported models share the plan quota, with 5-hour and weekly limits; quota units are not message counts.
- Pay-as-you-go API billing: use a separate API billing key and balance. Check the channel catalogue for model availability, supported protocols, and current channel pricing.
- Credit and quota are separate: topping up API credit does not increase subscription quota.
A subscription listing establishes inclusion, not successful execution of a particular tool workflow. Likewise, client support for a protocol does not mean every model is included in a subscription. Check your selected model, key type, and protocol as separate items.
Subscription availability and performance can vary with upstream load and policy changes. Review the current plans page before choosing that billing mode for a workload with strict stability requirements.
Configure Claude Code with the Documented Base URL
The Claude Code setup guide uses the Anthropic protocol, ANTHROPIC_AUTH_TOKEN, and https://llmapi.pro without /v1. Use the key that matches your billing mode. Replace YOUR_API_KEY locally; do not share the resulting credential.
macOS and Linux: existing Opus 5 configuration
export ANTHROPIC_BASE_URL="https://llmapi.pro"
export ANTHROPIC_AUTH_TOKEN="YOUR_API_KEY"
claude --model claude-opus-5
Windows PowerShell: existing Opus 5 configuration
$env:ANTHROPIC_BASE_URL = "https://llmapi.pro"
$env:ANTHROPIC_AUTH_TOKEN = "YOUR_API_KEY"
claude --model claude-opus-5
After checking current availability and configuring the matching key, select Opus 5.5 with:
claude --model claude-opus-5-5
The Opus 5.5 command combines the documented launch-flag pattern with the exact model ID listed on the subscription page. It is a configuration example, not a report of a successful request or a compatibility test.
Do not use the built-in /model menu as the site's availability list. The setup guide describes that menu as a client-side list unrelated to the model catalogue; the explicit launch flag is the documented selection pattern.
Audit the Four Request-Breaking Changes
Anthropic identifies these changes for code already running on Opus 5. Review each before replacing the model field in an API integration. The checks below are suggested application reviews, not observations of llmapi.pro behavior.
| Change | What to review |
|---|---|
| Thinking cannot be disabled | Locate requests that attempt to turn thinking off. Do not carry that setting into an Opus 5.5 request unchanged. |
| Forced tool use returns an error | Identify workflows that require forced invocation. Evaluate a compatible tool-selection approach before migration. |
| Thinking blocks are tied to the model and conversation | Review stored histories, block replay, and cross-model reuse. Do not assume thinking blocks are portable between models or conversations. |
computer_20251124 is not accepted on specified official platforms |
Check the configured computer-use tool version and the platform receiving the request. |
Always-on thinking and effort
Opus 5.5 uses always-on adaptive thinking and defaults to medium effort. Anthropic documents effort as the control for thinking depth; it is not an off switch for thinking. Review both your explicit request settings and any defaults inserted by your client.
For implementation details, use the effort and adaptive-thinking references accessible from the official Opus 5.5 page. Do not guess an effort field's placement in a request body.
Keep the computer-use restriction platform-specific
The documented rejection of computer_20251124 applies to the Claude API and Claude Models on Google Cloud (Google managed). That wording is narrower than a restriction covering every Google Cloud offering or every intermediary.
Do not interpret it as either a verified llmapi.pro rejection or a claim of llmapi.pro support. If your application uses computer tools, include the exact tool version and selected protocol in your account-specific compatibility check.
Diagnose Missing Progress Text Separately from Request Failure
The inter-tool text change is different from the four breaking changes above. Anthropic says text between tool calls is returned in thinking blocks whose text is empty at the default display setting. Applications streaming that text as progress updates go quiet between calls until they select a display value that returns it.
For a progress interface, review both the response consumer and the presentation layer:
- Identify which response blocks currently supply progress messages.
- Check whether the interface expects ordinary text between tool calls.
- Run a representative tool workflow and examine progress visibility as well as the final result.
- Consult the official feature references for a display setting that returns the text, and verify compatibility with the selected service and client.
These are suggested validation steps, not a tested payload recipe. A missing inter-tool message alone should not be treated as proof that the request failed: the official documentation explicitly describes this as a response-shape change that does not fail requests.
Separate Official Limits from Your Service Configuration
The official model overview and Opus 5.5 page document these vendor specifications:
| Item | Official Opus 5.5 specification |
|---|---|
| Claude API model ID | claude-opus-5-5 |
| Context window | 1M tokens |
| Maximum output | 128K tokens |
| Thinking | Adaptive, always on |
| Default effort | medium |
| Input → output | Text and images → text |
The official page also documents up to 300K output tokens on the Message Batches API with the output-300k-2026-03-24 beta header. That is a batch-specific capability, not the ordinary maximum output and not confirmation of llmapi.pro support.
Use the official specifications to frame your evaluation, then check the limits and features available through your selected billing mode and protocol. A model appearing in a catalogue does not by itself verify all of its vendor-documented capabilities through that service.
Migration Checklist and Current Pricing
Before replacing an existing Opus 5 configuration, use a small set of representative tasks to check:
- The exact model ID, current inclusion, matching key, and client protocol.
- Requests that disable thinking or force tool invocation.
- History handling for thinking blocks tied to a model and conversation.
- Computer-use tool version and the applicable platform restriction.
- Progress text between tool calls, separately from the final answer.
- Intended effort and output limits for the workload.
These checks are an evaluation plan, not a benchmark or a promise of improved results.
For current subscription terms and inclusion, use the plans page. For metered API prices and protocol choices, use the channel catalogue. Official vendor pricing is listed on the Opus 5.5 page; keep it separate from llmapi.pro's subscription and channel pricing.
The practical migration is more than selecting claude-opus-5-5: verify access, remove incompatible request assumptions, preserve conversation boundaries, and check progress visibility before relying on the new configuration.