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

A screenshot shows Gemini 3.7 and 3.6 Flash and 3.1 Pro all marked Leaving Soon, with Claude and GPT plans also expiring, leaving only Gemini 3.8 Flash.

Interface references in Antigravity show Gemini 4 Argon offering 256K, 512K and 900K context tiers, with larger context using more quota; no public release is confirmed.

A leak says Google is testing a model called Gemini 4 Carbon whose coding ability reaches the level of Opus 5.5, and the poster jokes that Argon may have been forgotten.

A Google email announced that Gemini 3.7 Flash is being retired in AI Studio in favor of 3.8 Flash, and the poster complains that Gemini 4 still has not arrived.

A video proposes a split coding workflow in which Gemini 4 Argon handles repo context, GPT-6.1 Sol writes patches and Claude Opus 5.5 does adversarial review.

A post jokes about Gemini 4's internal codenames running Argon, Barium, Carbon in alphabetical order, saying Carbon is already being tested internally while Argon is still not public.

A user asks about Gemini 4 Flash and suggests that Gemini 4 Argon ultimately ended up merged into the Flash line.

An emotional cheerleading post says Gemini 4 Argon is a beast still in its cage and will leave doubters in despair once released, with no model or benchmark details.

A user notices the tiered option for Gemini 3.7 Flash has been removed from Antigravity and reads it as a sign of making room for Gemini 4.

A satirical post notes that Gemini 4 Argon has not shipped while Gemini 4 Carbon is already in testing, mocking Google for keeping products ahead without letting users try them.

A rumor says Argon first reached Fairwind cyber-defense testers and a full rollout may follow, with early tests claiming strong complex-task performance and possible gains over GPT-6 Astra, none of it confirmed.

The Spanish-language podcast MonosPodcast devotes an episode to Gemini 4 Argon, gathering everything currently known about the model ahead of hands-on testing.