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

A user speculates Gemini 4 Flash is close to launch with a model codenamed Carbon behind it, arguing it would be an interesting release if it repeats the 3.8 Flash jump and fixes pricing.

Users note Gemini 4 Argon holds the highest humor score on benchmarks, and its actual replies back it up as the funniest model around.

A leak says Gemini 4 Flash has appeared internally under the codename Hydrogen, with no sign of Gemini 4 Carbon and speculation that the Arena Argon checkpoint was actually Flash.

Ad-funded app freebuff claims no-card, no-credit access to several models including one Gemini 4 Argon session per day.

Rumor says Gemini Carbon, a new Gemini 4 model built on Argon, is in early access for Google teams and partners, targeting long-horizon coding, agent workflows and 3D generation at 2–4x slower speed.

On day one of an Auto Fix experiment, the author had Gemini 4 Argon keep only production-code logic errors and fix them, with better results than expected.

Users report Gemini 4 Argon is gradually rolling out in Antigravity IDE, with the poster surprised it landed later than expected.

A post claims Kimi K3.1 Pro beats Gemini 4 Argon by 5 points on coding cost-performance and notes Argon's discounted pricing ends November 30, rising to $4/$10.

A leak says Gemini 4 Flash, codenamed Carbon, is launching soon and already looks stronger than GPT-6.1 Terra and Argon on coding and 3D, with a guessed price of roughly $0.50–$0.75/$2–$4.

The author complains that Google routes requests to different Gemini 4 Argon tiers based on subscription level, and pans GPT 6.1 Terra as weaker than Sonnet 5.5.

A user asks whether anyone has actually migrated from Claude or ChatGPT to Gemini 4 Argon since its release, inviting real switching experiences.

A leak claims the Flash model in the Gemini 4 family carries the internal codename Hydrogen, with no specs or launch timing yet.