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

A screenshot suggests Gemini 4 Argon is now available to paid AI Pro subscribers in Antigravity, a sign of a gradual rollout.

A Japanese user reports that Gemini Argon's quota drains at a rate comparable to Astra and Fable, and also complains about its hallucinations.

A French creator posts a video comparing Gemini 4 Argon with Claude, noting they checked who ran the benchmarks and urging viewers to try both on their own projects.

All three models rebuilt the same game visuals for Royale: Last Agent Standing inside OpenRouter's Ori Harness, which OpenRouter used to promote its 500+ model catalog.

A Google Creative Lab member describes collaborating with a 3D artist who uses Gemini 4 Argon to create 3D pieces.

Google Cloud announced that its new frontier model, Gemini 4 Argon, is now available, with a video walking through 10 internal demos from legacy code migration to autonomous Android UI.

The author shares a Gemini Argon image titled "ABYSSAL UNDERCROFT // QUALIA FRACTIONATOR & ANANKE ENGINE."

theo questions a leaderboard result where Gemini 4 Argon ties Opus 5.5 and Sonnet 8 sits ahead of Opus, calling it garbage.

Comparing Google's official Argon 3D demo with a version Opus 5.5 made in under a day, the author says Argon still lags clearly on realism and object placement.

Vals AI's Robo-Bench has frontier models control a Unitree G1 humanoid in real time; Gemini 4 Argon scored 0.83%, and most models failed by thinking too long to send commands.

On the same 3D cat-petting simulator task, Gemini 4 Argon took 3h22m and $30.50 versus GPT-6 Astra's 1h10m and $29, though the author found Gemini's cat more realistic.

A user's tests found Gemini 4 Argon's Turkish performance disappointing, with no clear gain over 3.1 Pro; it still beats most other models but shows no meaningful progress.