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

LuminaBench posted that it would love to see Gemini 4 Flash soon and joked about putting RSI to use.

A Gemini 4 Argon hackathon with the DeepMind team is set for next weekend at SHACK15 in San Francisco, with $20,000 in cash prizes and applications now open.

A user who tested Gemini 4 Argon before release says Google did not nerf the model after launch and that its performance is even better than before.

An Arabic post says uploading a PDF lets Canvas generate a visual interface in seconds, turning charts and text into an explorable dynamic experience.

The author fed the original post's video to Gemini 4 Argon and asked for a similar interactive depth-of-field demo; it was built fully in three.js in 45 minutes and is live to try.

The author lists free APIs including OpenCode Zen, River AI and Beam, and notes Gemini 4 Argon is paid-only in the Pro plan with no free API tier.

The author used three.js and Gemini 4 Argon to build an interactive synovial hydrodynamics and knee joint rheology demo, praising Argon's depth on non-Newtonian fluids and intra-articular pressure.

A post notes that Google AI Pro subscribers trying Gemini 4 Argon in AI Studio can set up a $5 prepaid billing balance to claim a $10 monthly API credit, with a link to the official details.

A Google developer account shows how Gemini 4 Argon, with Antigravity and Blender MCP, turns fragmented archaeological data into explorable 3D reconstructions of ancient ruins.

A user who mainly relies on GPT-6.1 Sol says Gemini 4 is currently expensive with unclear caching, treats it as a backup model like Grok, and looks forward to Gemini 4 Flash.

A user reports that Gemini 4 Argon is now unlocked for standard Google AI Pro subscribers, no longer limited to higher tiers or a gradual rollout list.

A Google account demo shows Gemini 4 Argon's Canvas converting dense PDFs into interactive, animated web apps that users can explore at their own pace.