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

The updated AA Intelligence Index puts Kimi-K3.1-Pro fourth at 55, reportedly ahead of Claude Fable 5.1, GPT-6 Astra and Gemini 4 Argon.

Arena tested Kimi K3.1 Pro on 3D scenes, HTML animations, websites and UI builds, comparing it with Gemini 4 Argon, Opus 5.5 and GPT-6.1 Sol.

Kimi K3.1 Pro climbed to fourth on the AA Intelligence Index, behind Opus 5.5, Sonnet 5.6 and Muse 4 but ahead of GPT-6 Astra and Gemini 4 Argon.

A user surfaces a googleapis/python-genai commit from four months before 3.6 Flash that references "gemini-4-flash-preview" and "gemma4".

A user criticizes OpenAI after comparisons put Opus 5.5, Gemini 4 Argon and even Kimi K3.1 Pro ahead of GPT-6 Astra.

A post claims the open-weight Kimi K3.1 Pro outperforms GPT-6 Astra, Gemini 4 Argon, and Muse Wonder in comparisons.

The author tried a mechanical pelican-on-a-bike demo with Kimi K3.1 Pro in Kimi Desktop and found it roughly on par with Gemini 4 Argon, minus the nitro boost effect.

A user asks how good Kimi K3.1 really is versus Gemini Argon, noting Argon is missing from the benchmark listing.

A developer describes using Antigravity with Gemini 4 Argon alongside stop motion, hardware and 3D modeling to build physical projects.

TheZvi walks through how to interpret benchmark graphs, using the Gemini 4 Argon and Opus 5.5 model cards as examples.

A user shares a video demo of the kanji "Crow" generated with a suspected Gemini 4 checkpoint, Carbon or Argon.

The poster lists recent Chinese model releases from DeepSeek, MiniMax and Kimi while noting Gemini 4 Flash and other US models are still awaited.