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

A user shared a pagoda image generated by Gemini Argon, calling it the most realistic version they have ever seen.

Gemini 4 Argon ties GPT-6 Astra at 53 on the Artificial Analysis Intelligence Index at about 60% of the cost, with strong agentic task gains and the lowest hallucination rate among top models, priced at $2/$10 per million tokens with 50% off until Nov 30.

A comparison image shows the frontend output of Gemini 4 Argon High generated with skills versus without skills.

A second Japanese post from the same period, with an image, says Gemini 4 Argon has arrived, and can be compared with the earlier post to see if it is the same entry point.

A user reports that AI Pro members can now use Gemini 4 Argon, but the quota is consumed very quickly, and warns others not to get too excited.

A Japanese user says Gemini 4 Argon seems smart but isn't enough to justify signing up for an Ultra plan in Antigravity.

A reply remarks that Argon did surprisingly well this time and asks whether it received a capability upgrade.

A Spanish-speaking user says it is getting harder to show a new model's advantages and points to Ammaar's thread on what has been built with Gemini 4 Argon.

A user says making Gemini 4 Argon Ultra exclusive to Antigravity after months of delays is disappointing, and that Ultra subscribers should get higher usage limits instead of exclusive model access.

The poster gave Gemini 4 Argon and Opus 5.5 the same prompt for a voxel-style pelican doing a wheelie on a flying fighter jet in a single HTML file, and opened a poll on which wins.

Gemini 4 Argon scores 84.1 on LiveBench 2.0 for fifth place, leading reasoning and data analysis but ranking last among the top five on agentic tasks and design, at $1.83 per task.

A user says Gemini 4 Argon looks insanely good at 3D modeling, without giving test details or comparison data.