Gemini 4 Pro / Flash: benchmarks, hands-on tests and leaks

A user says one Argon prompt used up 39% of their quota and that Flash also drains usage faster than before, suspecting a usage-accounting bug or backend change.

MathArena has released three new math benchmarks, ArXivMath, BrokenArXiv and ArXivLean, along with evaluation results for Gemini 4 Argon.

A comparison says building the same Eiffel Tower took Claude Opus 5.5 $33.45 and two hours, while Gemini 4 Argon did it in one shot for 23 cents.

Antigravity removed Gemini 3.1 Pro (High) from AI Pro in favor of Flash, Gemini 4 Argon now requires the $100 Ultra plan, and the $5 AI Plus tier loses manual Pro model selection on November 13, drawing complaints on the official forum.

One user says upgrading to Ultra for Gemini 4 Argon isn't worth it, calling the usage limits a joke and telling Pro users not to do it.

Palantir added four new models to AIP on October 15: OpenAI GPT-6.1 Sol, Anthropic Claude Haiku 5.5, Google Gemini 4 Argon (supporting IL2–5 environments) and GLM-5.3.

A post shares a link and steps to claim a free one-month Google AI Pro trial, saying it unlocks Gemini 4 Argon and premium AI features and can be canceled anytime.

The October 16 evening news roundup says Gemini 4 Argon has expanded from Ultra-only to Google AI Pro, with a dedicated weekly Argon quota counted separately from Gemini 3.8 Flash.

Users report that Gemini 4 Argon is now available to Google AI Pro subscribers in Antigravity.

Argon expands from Ultra to the Pro plan in the Antigravity app and CLI, with a weekly quota separate from 3.8 Flash, aimed at complex coding and agentic tasks.

A second set of claim steps: open the Google One dynamic plans page, check whether your account has the free trial, activate it if eligible, and cancel anytime before the trial ends.

Developer sickdotdev posted a screenshot of Gemini 4 Argon's output and said the model is AGI.