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

The developer updated their drone racing game Heatsink using Gemini 4 Argon with Crayon and three.js, adding new modes, atmosphere, drone upgrades and aero kits.

A hands-on test finds Gemini 4 Argon noticeably better at reasoning and long-context memory with fast responses, usable with a Jio account, but currently locked to the web and not available in Antigravity.

A Japanese user posts images saying Gemini 4 Argon has arrived, showing the interface but not detailing its capabilities.

According to Reuters on October 13, Gemini 4 Argon costs $2 input and $10 output per million tokens, doubling to $4 and $20 after November 30, with a unified 1M context and Antigravity limited to the Ultra tier.

A comparison post argues that Opus 5.5 is clearly superior to Gemini 4 Argon, without giving any specific test details.

The author tests Argon across coding, writing, design and knowledge work, finding the model back at the frontier but not ahead of rivals, while its Antigravity app trails competitors by six months to a year.

A 10-minute video walks through Gemini 4 Argon's official benchmarks, real-world examples and pricing, plus five practical tips for using it.

A blogger says the Gemini Argon 4 launch pushed OpenAI to respond, with Tibo announcing ChatGPT will optimize local disk storage starting the next day.

A user's long-term test finds Argon catches misread lines, 1.3-second silences and mismatches between charts and narration, and is more accurate when audio and video are fed separately, but its output is long-winded and best used alongside Opus 5.5.

Google AI Pro and Ultra subscribers can use Argon in the Gemini app, developers can access it through the paid Gemini API, and the macOS app adds an Antigravity coding interface.

A user finds Argon uses far fewer tokens at minimal effort, but the gap in steps and token use from minimal to high is small, suggesting the lower effort levels offer little benefit.

A user is annoyed that Gemini 4 is only available to Ultra subscribers in Antigravity, questions where the TPU capacity went, and says the model was meant to be Google's answer to Opus 5.5.