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

On real-robot tasks Opus 5.5 and Gemini 4 Argon both score 48%, GPT-6 Astra 47%, while Kimi K3.1 Pro trails at 9%.

Some users spotted an anomaly in Google Antigravity's model list, prompting speculation that Gemini 4 Carbon, possibly alongside Gemini 4 Flash, is about to launch.

A user notices Gemini Flash and the latest Argon are absent from a set of benchmark results and asks whether the charts show the full picture or only flattering numbers.

A user jokes that the Gemini family is gathering to welcome "Uncle Carbon," saying Gemini 4 Carbon will bring along its nephew Gemini 4 Flash, hinting both models are imminent.

A user reports the Gemini 4 Argon quota has been reset and plans to measure how many more tokens it allows, hearing the limit may have doubled.

On the same Eiffel Tower replication task, Kimi K3.1 cost $2.04 in 8 minutes versus Gemini 4 Argon's $6.10 in 9 minutes, with Kimi judged cheapest and most accurate.

A developer used Gemini 4 Argon to build Kinetic3D Studio, a single HTML file that turns 2D screenshots into 3D promo videos locally in the browser via WebGL and Three.js, with no backend.

A user says Jio Pro is supposed to unlock Gemini 4 Argon and Opus 5.5 in Antigravity, but it still doesn't work for them and asks if anyone has access.

A post claims Kimi 3.1 Pro trails Opus 5.5 by about 3 points on several intelligence indexes while overtaking GPT-6 Astra and Gemini 4 Argon, with open weights due Nov 4.

HarshithLucky3 prompted the Gemini 4 Carbon checkpoint in Arena for a random 30-second late-night jungle horror scene, saying its sound design beats even Argon.

A developer shares a screenshot claiming Google is preparing new models named Gemini 4 Flash and the previously unheard-of Gemini 4 Carbon.

A promotional post praises Gemini 4 Argon's 3D capabilities and long-context handling of large projects while plugging a cheap Google AI Pro subscription.