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

A user had Gemini 4 Argon High generate an 11-layer riding-tower SVG with a pelican on a bicycle, flamingo, turtle and a cat leading the stack.

A Pixel 11 Pro owner says the bundled Gemini Pro plan excludes Gemini 4 in Antigravity and that Gemini 3.1 access was also removed, calling Google dishonest.

A post aimed at Claude users asks them to name one reason not to switch to Gemini now that Gemini 4 Argon is out.

The author used Gemini 4 Argon to repurpose a long video into a 9:16 short; it worked almost in one shot, needing only a 'make it under 1 minute' revision.

A short reply asking for Gemini 4 Flash, with no additional details.

An AI weekly report covers Claude Haiku 5.5, multi-agent orchestration, GPT-6.1 Sol and Meta Muse, and includes Gemini 4 Argon though its section is cut off.

A Japanese poster compares Gemini 4 Argon with Claude and ChatGPT and concludes it is stronger than expected, with the test video linked in a reply.

Instead of burning tokens animating humanoids, the author generates a raw mesh in Tripo or Meshy and has an LLM such as Sol, Astra, Argon or Opus guide Mixamo rigging, yielding a rigged animated 3D model in about ten minutes.

In coding benchmarks DeepSeek V4.1 Flash-1017 matches Gemini 4 Argon on some tasks at far lower cost than GPT-6.1 Sol, though still behind Claude Opus 5.5.

A user who accessed Gemini 4 Argon through Perplexity says it rivals or beats OpenAI and Anthropic, and asks whether Google's Ultra plan is worth buying.

One-shot task took Gemini 4 Argon about 10 minutes and ~2% of the author's weekly quota, versus ~4 minutes free on Open Code for DeepSeek V4.1 Flash, which the author still rates below Argon.

A user paired Hermes Agent with Gemini 4 Argon to design a 3D-printable case for a dual-DGX setup, with the model loaded into a slicer rather than being a concept render.