Google Antigravity refreshed its model lineup on October 3, 2026. Google updated its official model documentation to list Anthropic's Claude Sonnet 5.5 (thinking) and Claude Opus 5.5 (thinking) as selectable reasoning models, so developers no longer need a separate Anthropic API key to run Claude's newest models inside Google's agentic coding platform. The rollout comes with a clear paywall, though: only eligible Google AI Pro and Google AI Ultra subscribers can select the two models.
Free, Plus and trial Pro accounts are not on the list
According to the official model documentation as checked on October 3, 2026, eligibility for the two Claude 5.5 models is spelled out in detail. A Google AI Pro subscription must be a full paid subscription; Pro accounts on a promotional trial do not qualify. Free accounts, Google AI Plus and Enterprise accounts are not listed as eligible at all. Antigravity itself now shows a matching notice stating that Opus 5.5 is available on paid Pro and Ultra plans, and warning that third-party model access will no longer be available on the current plan starting November 2, 2026. In other words, third-party models are being withdrawn from Antigravity's free tier, and Claude 5.5 is simply the first batch restricted to paying users. In the same documentation, Gemini 3.8 Flash, Gemini 3.7 Flash, Gemini 3.6 Flash and Gemini 3.1 Pro remain available across all listed plans, so the gap between first-party and third-party models is now written directly into the product rules.
Three older models retire on November 2
Confirmed alongside the launch is a retirement list: Claude Sonnet 4.6, Claude Opus 4.6 and the open-weight model GPT-OSS-120B will be removed from Antigravity on November 2, 2026. Those models are still selectable for now — the two Claude 4.6 models are available to Free, Plus and Pro users, and GPT-OSS-120B also includes Ultra — but the countdown has started. Teams still running fixed workflows on these models, or holding saved default model configurations, need to finish switching and regression testing before November 2; waiting until removal day means automated jobs lose their model outright. One easy detail to miss: the documentation says a reasoning-model selection persists within a single conversation, so switching models while an agent is running does not change the model used for the active turn — you need a new conversation to change models.
What this change actually means
On the surface, a coding tool simply gained another set of model options. One level down, it turns "using Claude inside Google's tool" from a trial perk into a subscription entitlement. For developers weighing Antigravity against Anthropic's own tooling, the math changes: Antigravity lets Gemini and Claude sit in one agentic environment billed against subscription quota, with no separate API billing to maintain; the trade-off is that model availability is entirely governed by Google's plan rules, and this time Enterprise is not even in the first batch — a reminder that the eligibility table cannot be taken for granted. A concrete step for teams on Antigravity: this week, check your account's plan type and trial status first, confirm whether Claude 5.5 is actually selectable, and only then decide whether to move core workflows over; at the same time, treat the November 2 removal of the older models as a hard deadline and update saved configurations early.
