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

A user shares their first test output from Argon, an image of a pelican doing a wheelie on a fighter jet, and calls the result insane.

A leak claims Gemini 4 Argon is now live on Google Vertex AI, with a screenshot attached.

A user says Opus 5.5, GPT-6 Astra and Gemini 4 Argon are good at 3D modeling and building scenes, but still struggle with character rigging and sculpting.

AiHubMix ran the same prompt through its API on both models, reporting Argon finished in 3 minutes for $0.50 while Astra took 25 minutes and $11, and said Argon is now live on its platform via Google Vertex AI.

Someone spotted traces of Anthropic's Fable 5.5 and speculated it could launch the same day Google DeepMind releases Gemini 4 Argon.

TokenGremlin says Argon is already usable on some platforms with a Google announcement imminent, while Anthropic's Fable 5.5 is also starting to appear in provider configurations.

A leak says Argon has gone live on partner APIs ahead of Google's own services, with comparisons against Opus 5.5 and Meta Muse Wonder expected soon.

A user says Gemini 4 Argon is available on third-party relay service ZenMux rather than through Google, and questions claims that relays run models locally.

User argofowl says a tweet looked like it was published early, which they read as a sign that Gemini 4 Argon is about to launch.

A user noticed that Gemini 4 Argon appears to be live, but could not find any official release announcement from Google.

A user ran the "Otter on a Plane" prompt on Gemini 4 Argon at default effort, getting 15,514 output tokens (2,145 of them thinking) in 112 seconds for $0.155.

A user tested Gemini 4 Argon at high effort on a pelican SVG, producing 19,809 output tokens (6,289 of them thinking) in 143 seconds for $0.198 at $2/$10 per 1M tokens.