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

The post says signing in to v0.app with a Google account gives monthly free credits to try Gemini 4 Argon, with no payment or card required.

AI Studio lead Logan says Argon's compute cost is too high to open to everyone, and says a Gemini 4 version for billions of people is in the works, hinting at a Flash release.

A long-time Gemini 4 user says the model has a unique style in writing, code taste and design color choices that sets it apart from other frontier models.

A Google DeepMind blog post shows Gemini 4 Argon use cases across the robotics stack: agentic reasoning, real2sim, control policies, value prediction and progress tracking.

A short post with an image in which the author announces they have been given access to Gemini 4 Argon, a signal of a limited or gradual rollout.

A user compared Gemini 4 Argon and Opus 5.5 on the same prompt, finding the quality similar with Opus slightly better, and was surprised Google closed the gap so much.

Gemini 4 Argon was tested against Gemini 3.1 Pro on real multi-step enterprise document tasks, scoring 85% vs 71% overall and leading by 23-30 points in healthcare, life sciences and data analysis.

A comparison video says Gemini 4 Argon beats Muse Wonder on a 3D physics simulation test, without giving test details or scoring criteria.

On the same task, Argon 4 scored 8.4 at a cost of $3.42, below Opus 5.5's 9.1 and Sonnet 5.5's 8.5 but above GPT-6 Astra's 8.0, making it strong on value.

A user says Gemini 4 Argon built CAD models from scratch without importing any files, including a tendon-driven robot hand.

A user reports that hands-on use of Gemini 4 Argon in Antigravity shows it is not optimized only for benchmarks.

A user shares artwork made entirely in code with Gemini 4 Argon and tags a collaborator.