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

A developer says they built an entire web app with Gemini 4 Argon without writing a single line of code and shared a playable link.

A user had the openclaw agent complete a Google One subscription on their Gmail account, with Google AI Pro active 3 minutes 54 seconds later, then started trying Gemini 4 Argon.

Gemini 4 Argon (high) enters the Debate Benchmark at #4 with 1676 points, 190 above Gemini 3.1 Pro Preview, while Meta Muse Wonder 1.0 (high) debuts at #5 with 1655.

A developer says they love using Gemini 4 Argon to refactor and performance-optimize codebases, and notes how much models improved and dropped in cost in six months.

A developer devirtualized Easy Anti-Cheat for Linux using Gemini 4 Argon and GPT-6.1 Sol through Hex-Rays IDA Assist's MCP, with the code open-sourced on GitHub.

A developer built a keyboard-navigable 3D farm with Gemini 4 Argon and Antigravity where animals tell first-person fun facts, using Nano Banana for art, Blender for models, Gemini TTS for voices and Lyria for music.

A post says Google engineers are using Gemini 4 Argon to untangle Chromium's legacy code and make Gemma models nearly three times faster.

A comparison post lines up Gemini 4 Argon against GPT-6.1 Sol, Claude Opus 5.5 and Grok 4.7 and asks which is the real coding beast.

The author notes that Gemini 4 Argon's benchmark announcement lists input at $2 per million tokens and output at $10 until November 30, rising to $4 and $20 from December 1.

A video shows a water physics simulation generated by Gemini 4 Argon, which the poster calls another level.

A complaint post says Argon, Kimi and Muse now look like jokes and argues Anthropic could win the frontier model race by releasing Fable 5.5 today.

A highly liked post says only that "Gemini 4 Argon is AGI," with an image showing its performance, a typical first-impression reaction.