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

A user who made 3D content with Gemini 4 Argon found the results acceptable but still mediocre compared with Claude's models.

Argon scored 87% on the ERQA robot spatial reasoning benchmark, covering robot control, Real2Sim, physical understanding and self-improvement.

A Chinese blogger says Gemini 4 Argon is available today in AI Studio, the Gemini API and the Gemini App, with no official benchmarks and a focus on physical-world understanding and engineering demos.

Developer Angaisb added Gemini 4 Argon to his SubChat site, letting subscribers test it without buying a Gemini subscription, with $35 in API usage included.

After early access, Mollick says Gemini 4 is a big jump over earlier Gemini models with a more pleasant, more agentic personality, but still lags at complex multi-agent work and is held back by the Antigravity harness.

A short post arguing against betting against Google in the AI era and saying Gemini 4 Argon looks great and is worth trying.

A video shows a black hole generated by Gemini 4 Argon, with the author noting it used only one Python file and no JavaScript or shaders.

A user shares how to sign up for v0.app with a Google account and use its monthly credits to try Gemini 4 Argon for free.

A designer used Gemini 4 Argon in Antigravity to prototype a playable WebGL mahjong game with ceramic tiles, sound effects and spring physics.

A user jokes that Gemini 4 Argon can generate enormous voxel pagodas, but building them at that scale would require six or seven Ultra subscriptions for enough quota.

zacharynado shares the official Google blog link announcing Gemini 4 Argon and invites people to share their experience with the model.

A user posts an image joking that "Hakimi is back" and says they envy those who already have access to Gemini 4 Argon.