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

A weekly AI recap video rounds up new model news including Gemini 4 Argon, GPT 6.1 Sol, Claude Sonnet 5.5 and more.

A poster says Gemini 4 Argon is "preparing to sortie" and, though the cited Gemini API doc couldn't be found, thinks it could arrive within 24 hours.

Google's Arabic blog says Argon is rolling out to paid API customers and AI Ultra users, but API tests still return 404 and public access is unconfirmed.

A leak claims Gemini 4 Pro has a release date of October 9, with no official confirmation.

The author says Opus 5.5 still has an "AI feel" in Japanese, still uses Gemini 3.8 Flash for writing, and hopes Gemini 4 Argon will offer both.

An unsupported post argues that Gemini 4 Argon has finally surpassed GPT 6 Astra and asks others to change the author's mind.

A user says Google has surpassed OpenAI on VAL's RSI index, speculates about the next Pro model if Gemini 4 Argon is behind it, and notes Google confirmed higher internal checkpoints than Argon.

A user says a friend's Antigravity model picker suddenly showed Gemini 4 Argon, which he used for a while, suggesting Google may be quietly rolling it out to Ultra users.

The poster says they don't care which GPT and Claude models are in Antigravity and just want Gemini 4 Argon and 4.0 Flash to show up in Google AI Studio and the Gemini App one morning in early October.

On the AA Intelligence Index, Gemini 4 Argon (High) scores 53, level with GPT-6 Astra (Max) and below Claude Opus 5.5's 58, at $2/$10 per million tokens, a fifth of Astra's price.

A long analysis says Google's Antigravity coding tool offers Claude Opus 5.5 as its top model with opaque quotas, while Gemini 4 Argon scores well but is called weak at coding and limited to a few partners.

A user notices that Gemini 3.6 and 3.7 Flash in Antigravity are about to be removed and guesses two Gemini 4 models will take their place.