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

A poll post asks readers to pick the best model for writing real production code among Gemini 4 Argon, GPT-6 Astra, Grok 4.7 and Claude Haiku 5.5.

A user says they fear Gemini 4 Argon may be stronger than OpenAI's Astra and argues OpenAI should release new models instead of resets.

A post explains using Token monitor to check if Antigravity routes to the Gemini 3.8-flash-exp-a model, saying 2 of 7 AI Pro accounts had it rolled out.

A post jokes that Argon hype is dead and Carbon is next, guessing Gemini 5 Krypton and Silicon may follow.

A user says their main account did not get the Argon rollout but a family group sub-account did, similar to Opus 5.5, and suggests Antigravity and AI Pro users try it.

An account posted an image with the caption "👀", hinting at news related to Gemini 4 Argon.

A commenter argues the AI race is moving faster than Google can ship, so Opus 6 will be out by the time Argon arrives and Argon will be dead on arrival.

A joking post notes that Claude Opus 5.5, Gemini 4 Argon and Sonnet 5.5 all solve the problem, while choosing which model should do it takes 40 minutes.

A Chinese post with an image claims Google achieved absolute dominance in a benchmark, with Gemini 4 Argon far ahead and the other players' combined scores not enough to compete.

A poster says Anthropic answered Google's Gemini 4 Argon a few days later with Claude Haiku 5.5, described as its cheapest and fastest small model yet and better at computer use and agent tasks.

A poster claims Gemini 3.8 Flash is actually routing to the Argon codename, reportedly Gemini 4 Pro, and asks people to test it.

The post says 3.8 Flash is currently routed to Argon and that Gemini 3.1 Pro in Antigravity now points to gemini-3.8-flash-exp-a, suggesting Google is testing Gemini 4.