Cannot find Claude 5.5 in Antigravity? Check Pro trials and model access

Step 1: Record your exact account conditions

First check that the signed-in account is the one with the subscription, then record the full plan name and whether it is a trial. Do not simply say “I have already paid” when troubleshooting; that does not distinguish the different plans. This article is based on official documentation accessible on October 3, 2026, and does not treat that verification date as the feature's launch date.

Keep a record containing only the necessary details: client used, plan name, trial status, and time of the issue. If you use both personal and company accounts, first identify which account owns this task, then compare model lists so that repeated account switching does not obscure the original conditions.

Step 2: Select the model in the conversation and wait for the current turn to finish

The model selector is below the conversation input box. The selection carries over to subsequent messages in that conversation. If you switch models while a turn is running, that turn continues with the original model until it finishes or is canceled. This article covers Antigravity 2.0 only; do not apply this interface path to every IDE or CLI.

Let the current task finish normally, then select the target model and send a separate small request. Do not infer whether the model switched from the tone of its answer. Check the selector and the interface information for that task, and record the selection before the request starts. If the list differs from expectations, save a record of what you see before proceeding.

Step 3: If the model is selectable but cannot continue, check quota settings

Settings shows the base quota for each model. According to the official documentation, quota consumption depends on task workload rather than a fixed equal amount per message. Pro's five-hour replenishment is also subject to a weekly cap.

Pro and Ultra can use AI credits to pay for usage beyond the base quota. Setting AI Credit Overages to Never waits for quota to recover; Always uses credits after the base quota is exhausted. Buying credits pays for additional usage and does not imply eligibility for models that the plan does not include.

If you are only investigating an issue, first record the current settings and the exact error text. You do not need to enable extra spending just to try it. Record “not present in the list at all” separately from “insufficient quota after selection” so that an eligibility issue is not mistaken for a need to buy credits.

Step 4: Check execution with a small task that has clear acceptance criteria

For example, prepare a demonstration file without business data and ask the model only to explain a function's inputs and outputs without editing the file. Before running it, write down the expectations: it can read the specified file, discusses only that function, and makes no code changes. Check each point afterward instead of treating a “task completed” reply as proof of success.

After the first verification passes, try a small, clearly scoped edit and inspect the diff. If it fails, note whether the failure happened before execution or during it, and keep the shortest reproduction steps. This is troubleshooting advice, not a claim that this article has tested your account, nor a guarantee that a particular model will be faster or more accurate.

Existing workflows using older models: allow time for migration checks

The official Models page lists Claude Opus 4.6, Sonnet 4.6, and GPT-OSS-120b for removal on November 2, 2026. If you still depend on them, verify available replacement models in advance.

Choose one explanation task, one small edit, and one troubleshooting example from your daily work, and record whether each result meets your requirements. Compare using the same inputs and acceptance criteria. One successful demonstration by a new model does not establish that all older tasks can migrate directly. For production code, continue the project's existing checks and human review.