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

A short post asking if anyone is still using Gemini 4 Argon, reflecting community chatter about its staying power.

A Japanese user says a lyric motion video was produced entirely inside the Gemini (Argon) chat with no plugins or MCP, sharing a five-step workflow from reference hunting to storyboard and generation.

GoogleDevsIN says Gemini 4 Argon is built for complex multi-step engineering work, promising fewer bugs, faster builds and more time for system design.

A developer reports that once cache misses are factored in, Claude Sonnet 5.5 costs less than Gemini 4 Argon, contrary to expectations.

A user says Gemini 4 Argon nailed coding on day one but now loops and breaks things when fixing them, and role-play has become stiff.

A meme about cross-model handoff, framed as Claude Opus 5.5 picking up code that Gemini 4 Argon wrote.

Polymarket launched a new market betting on when Gemini Argon will next be released, reflecting community interest in the Gemini 4 rollout timeline.

From being written off to Argon sitting at the top, the author says Google AI's reversal deserves study.

A user complains that Gemini 4 Argon's quota is too small and already exhausted, publicly asking Logan Kilpatrick when it resets.

A video compares Gemini 4 Argon with Sonnet 5.5, GPT-6 Astra and Opus 5.5: Argon is 7x faster and 4x cheaper than Opus with strong quality, while Opus 5.5 still leads on raw performance.

A user reports Gemini 4 is close to or possibly better than Opus 5.5 on complex tasks, notably strong at security work, and far more capable than Gemini 3.8 Flash.

A user says they gave Gemini Argon a single prompt to improve their website and quit as soon as it started responding, without reporting the outcome.