Anthropic Claude 5.5 Requires Pro or Ultra in Antigravity

Developers using Claude 4.6 or GPT-OSS-120b in Google Antigravity need to choose replacements before November 2, 2026. Their options depend on their subscription: Anthropic’s Claude Sonnet 5.5 and Claude Opus 5.5 are listed for Google AI Ultra and non-trial Google AI Pro, but not for Free, Google AI Plus, or Enterprise.

Eligible Pro and Ultra subscribers can move to the newer Claude models. Free and Plus users who want to keep their existing plans will need to choose a listed Gemini model.

Google’s Antigravity model table documents the access restrictions and removal deadline. It does not establish when the newer models first became available or guarantee that they already appear in every eligible account.

Which Antigravity Plans Include Claude 5.5?

Google lists Claude Sonnet 5.5 (thinking) and Claude Opus 5.5 (thinking) with the same subscription requirements:

Google’s footnote limits Pro access to “for non-trial subscriptions only.” A Pro trial therefore does not establish eligibility for either Claude 5.5 model.

Although NokiaPowerUser’s coverage also describes promotional and discounted Pro subscriptions as excluded, the official table specifies non-trial subscriptions. It does not separately settle every discounted, promotional, or student subscription arrangement. Users with those subscriptions should check their account; a discount alone should not be assumed to disqualify them.

Enterprise is not simply an additional tier above Ultra with every model included. In the current table, it has access to the listed Gemini models but neither Claude 5.5 model.

Non-trial Pro is the lowest eligible plan listed for someone choosing a subscription specifically to use Claude 5.5 in Antigravity. Both eligible plans include both Claude models, so Ultra offers a larger usage allowance, not exclusive access to Opus.

These requirements apply to Antigravity. They do not describe Anthropic’s own subscriptions, API availability, or the model choices inside the consumer Gemini app.

Claude 4.6 and GPT-OSS Leave on November 2

Google marks three models for removal on November 2, 2026:

• Claude Sonnet 4.6 (thinking)
• Claude Opus 4.6 (thinking)
• GPT-OSS-120b

The deadline appears in a footnote attached to those models in the official table. It concerns their availability inside Antigravity, not a global shutdown across other services.

Their current access differs from the new Claude lineup. Sonnet 4.6 and Opus 4.6 are listed for Free, Plus, and Pro, but not Ultra or Enterprise. OpenAI’s GPT-OSS-120b is listed for Free, Plus, Pro, and Ultra, but not Enterprise.

Free and Plus users face more than a model-version update. Their plans currently list older Claude models and GPT-OSS, but do not list Claude 5.5. After the scheduled removal, the documented replacement choices on those same plans are Gemini models.

Google currently lists Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash, and Gemini 3.1 Pro across all four plan categories. None carries the November 2 removal footnote in the supplied table, though their present listing is no promise that Google will preserve the lineup indefinitely.

There is no documented one-for-one successor to GPT-OSS-120b in this change. Users relying on it should evaluate an eligible alternative instead of assuming Antigravity will replace it with another OpenAI model.

The documentation also leaves open what will happen to conversations that still have a retiring model selected. Planning around an automatic substitution would be risky.

Choose a Replacement Before Changing Your Default

Start with a subscription check, then evaluate the replacement on a small sample of work that matters to you.

Free and Plus users can test the listed Gemini models without upgrading solely for this transition. If retaining access to Claude is a requirement, the documented options are non-trial Pro or Ultra.

Eligible Pro users can evaluate either Claude 5.5 model as well as Gemini. Moving to Ultra is not required to make Sonnet 5.5 or Opus 5.5 eligible.

Ultra users relying on GPT-OSS-120b can evaluate Claude 5.5 or Gemini. The current table does not list Claude 4.6 for Ultra, so the older Claude models are not a documented intermediate fallback.

Enterprise users should work from the Gemini options listed for their plan. A personal Pro or Ultra entitlement should not be assumed to change an Enterprise environment’s model access.

Google says the reasoning-model selector sits below the conversation prompt box. The selected model persists between messages within that conversation.

Changing the selector while the agent is running does not redirect the current turn. Antigravity continues using the previously selected model until it finishes that turn’s steps, or until the user cancels the execution. Letting a turn finish before switching makes it easier to tell which model produced which result.

A compact migration check:

1. Identify affected workflows. Note where you deliberately choose Sonnet 4.6, Opus 4.6, or GPT-OSS-120b.
2. Confirm the replacement appears in your account. Subscription eligibility in a table is not a substitute for checking the selector.
3. Try representative tasks in a controlled setting. Use a familiar bug fix, a refactor with clear constraints, or a change backed by tests.
4. Inspect the results before adopting the replacement. Review code changes, test outcomes, instruction-following, and usage consumption.

These are evaluation suggestions, not results from hands-on testing. The supplied sources do not establish that Claude 5.5 or a particular Gemini model will outperform the retiring models on your repository.

Keep the task, starting code, and instructions consistent for a more useful comparison. A successful answer to one easy prompt says little about whether a replacement can handle the longer, more demanding jobs that made the previous model valuable.

Pro and Ultra Differ in Quota, Not Claude Eligibility

Access alone does not tell you how much work a subscription can support.

Google’s Antigravity plans documentation describes Ultra as providing the highest quota, refreshed every five hours, and the highest weekly rate limits. Pro receives a lower allowance, with five-hour refreshes until its weekly limit is reached. Users outside Pro and Ultra receive a quota refreshed weekly.

Google does not provide fixed prompt counts in the supplied documentation. It says usage limits correlate with the amount of work the agent performs: straightforward tasks can consume less quota, while complex tasks consume more. Counting prompts alone will not tell you whether a subscription fits your workload.

Confirm model eligibility first, then assess the allowance. Pro is sufficient for documented Claude 5.5 eligibility; Ultra may matter when usage capacity becomes the constraint. Neither should be described as unlimited Claude access.

Pro and Ultra users can also use purchased AI credits, or one-time promotional credits, for usage beyond the baseline quota. The AI Credit Overages setting controls whether credits are used automatically:

• Never waits for the baseline quota to refresh.
• Always uses credits after the baseline quota is exhausted, then switches back when that quota refreshes.

Overages follow Google’s stated Gemini Enterprise consumption pricing. The supplied evidence does not include the applicable numerical rates, so it does not support a reliable cost estimate for a particular migration workload.

Check that setting before running extensive replacement tests. Otherwise, an evaluation may continue beyond the included allowance when you expected it to stop.

Google’s plans page also says bring-your-own-key and bring-your-own-endpoint options are not supported for additional rate limits. An external API key is therefore not a documented way around the quota.

The Removal Date Is Clear; the Rollout Date Is Not

NokiaPowerUser publicly reported the access changes in its October 3, 2026 article. Google’s model table has no publication timestamp, so October 3 should not be presented as a verified deployment date or a precisely timed launch.

Google documents the new model eligibility and schedules the three older models for removal on November 2. Availability has not been independently tested for individual accounts in the supplied reporting.

Developers can plan around the removal deadline even without a verified rollout date. Verify your entitlement, select a replacement, and test it before the retiring option disappears. Preserving acceptable results and predictable usage is more valuable than choosing the newest model name without checking either.