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

A user shares an SVG version of the Mona Lisa generated by Gemini 4 Argon at the high setting, showing its vector drawing ability.

A short comment calling Gemini 4 Argon an exceptional scientist, which makes the user suspect Opus may have been holding back.

A short post saying Gemini 4 Argon is in the works, with a joke about mixing Gemini with Web3.

A user calls a Three.js 3D demo made with Argon and Nanobanana in a closed improvement loop the most impressive general demo they have seen.

A satirical post cites Gemini 4 Argon's safety report, joking that the alignment strategy is to keep writing papers listing famous examples of what an AI should not become.

AI researcher DKokotajlo posts a screenshot from the Gemini 4 Argon report, joking that the models are becoming more situationally aware.

A $396K Polymarket bet on Anthropic releasing a 5.2+ model before October 31 has the October 31 contract at 87%, as the author says Google shipped Gemini 4 Argon straight to production without a waitlist and rewrote the release playbook.

A post lists the available entry points for Gemini 4 Argon: the Gemini app (AI Pro and above), AI Studio, Google Cloud Agent Platform, Gemini Enterprise, BigQuery generative AI functions, and Code Mender.

A user built a voxel dinosaur island with Gemini 4 Argon, finding Qwen 3.8 Flash Next looked better but that Argon added more dinosaurs that roam around.

A demo shows Gemini 4 Argon using a single detailed workflow prompt in Antigravity to generate a Japanese temple complex in Blender, with animation and sound added afterward.

A user says Gemini 4 Argon became available overnight, and using it as an agent via subscription requires Google AI Ultra, though pay-as-you-go API access works.

This episode of the BTLG weekly live show discusses Gemini 4 Argon, Muse, memory features and the AI sector selloff.