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

Dot's 14-day changelog lists new model availability including Gemini 4 Argon, DeepSeek V4.1 Flash and Muse Wonder, plus GPU marketplace and staking progress.

A user runs a complex SVG prompt through Gemini 4 Argon High: a 12-layer riding tower of pelican, flamingo, turtle and crab characters, plus a bee chasing a cat.

The author says they built their first playable 3D shooter, Shadow Strike, with Gemini 4 Argon in Google AI Studio, playable on mobile and laptop, but hit the usage limit at prompt 2.

The author compared DeepSeek V4.1 Flash, Claude Opus 5.5 and Gemini 4 Argon across 160 code-generated paintings, calling DeepSeek impressive but noticing an interesting pattern.

The author benchmarks Claude Fable 5.1, GPT 6 Astra, GPT 6.1 Sol, Grok 4.7 and Gemini 4 Argon on a self-made KnotBench of eight tasks including running code, bug hunting and admitting no answer.

A Korean user reviews the AI tools he pays for, saying he has mostly dropped Gemini but plans to try Gemini 4 Argon for writing after hearing it writes well.

CommonStack now offers Sonnet 5.5 for fast, token-light daily tasks and Gemini 4 Argon for heavy coding, math and multi-step agent workloads, both with million-token context.

PromptLLM shares an image post showing Gemini 4 Argon's answer to the question of what keeps people stuck in life.

A demo claims Google Argon edited a video in under 10 minutes and asks whether users would still pick Claude Opus 5.5.

A developer claims Gemini 4 Argon inside Droid can build an interactive Seven Wonders of the World project in about ten minutes.

The author says Gemini 4 Argon is strong at Blender and is powering four explorable 3D demos, but rented GPUs still take hours to render a few frames, delaying releases.

A user shares an SVG animation from Gemini 4 Argon High showing a pelican on a bike carrying an 11-layer tower of scooters, flamingos and turtles, and plans a DeepSeek V4.1 comparison.