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

Adidotdev posts an image claiming Gemini 4 Argon has begun a gradual rollout to some users in the US.

The post argues Google has spent too long on Gemini 4 Argon, so it could be overshadowed by Anthropic's Fable 5.5 and a new OpenAI model as soon as it ships.

With the same SVG prompt, Argon gets the M4 CS proportions, badge text and spec line right, while Flash turns the blue car green and shows a different spec line.

An Arabic post says Google raised Gemini 4 Argon's single-pass thinking limit from 64K to 1M tokens, gave it first to Fairwind cybersecurity experts, and cites $2/$10 pricing and a Wiz vulnerability case.

A poster asks why Gemini 4 Argon still isn't generally available when models like GPT-6.1 Sol and Opus 5.5 launched directly, and says Argon is clearly worse than both.

A post lists Google's moves after announcing Gemini 4 Argon only to Fairwind testers: cutting free/AI Plus access on Oct 9, releasing Nano Banana 2.1 and EmbeddingGemma 2, with Argon still not public.

A post rounds up rumors that Gemini 4 Argon may launch within a week, while some who use it say it seems a bit silly and may not measure up to Opus and Sol.

A user on a Chinese site says Gemini 4 Argon showed up in their model picker on the Gemini iOS app, which if true could mean a rollout within days.

A Korean-speaking user asks whether Gemini 4 Argon is available only in Spain, suggesting an unusually narrow rollout for the model.

Someone spotted the parameter argon_limit_reached in a Google upgrade link and speculates Gemini 4 Argon may soon be offered to AI Ultra subscribers.

A rundown of Argon from its September 30 announcement to October 6 access limited to Fairwind trusted testers, with no dates yet for the public API or the Gemini app.

A user asks when Gemini 4 Argon will actually reach the public, noting it has appeared in scattered places but has no release update almost a week after its official announcement.