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

A user describes juggling tools daily: Gemini Argon for code, ChatGPT and Claude for editing, Grok Bot for agents, and Instinct as a personal agent.

Google DeepMind showed Gemini 4 Argon building an interactive 3D anatomy model from open-source data, letting users toggle skeleton, muscle and organ layers and zoom into the heart, though it is an educational demo rather than a clinical simulation.

A post asks readers to choose between Opus 5.5 and Gemini 4 Argon as the single model their next 12 months would depend on.

A video shows Gemini 4 Argon producing a result from a single prompt, with the full prompt promised on jiro.build next Monday.

A user describes using Gemini 4 Argon to draft plans and passing them to Opus for review, advocating a second model as a checker.

Given a still and reference videos in Antigravity, Gemini 4 Argon drew and animated every bone, camera move and particle from scratch in Python without extracting assets; Opus failed first.

A user posts an image and asks whether Gemini 4 Argon is genuinely good or merely overhyped, sparking debate on its real-world performance.

On the same flight simulation task, Gemini 4 Argon finished in 23 minutes for $7.84, Opus 5.5 took nearly 3 hours and $31.85, and GPT-6 Astra landed in between.

A developer built an Apple-style website with Gemini 4 Argon, including custom 3D HomePod models so faithful the creator jokes even Apple lacks them.

A short reply says the user runs Claude and ChatGPT but finds Gemini's Argon model good too.

A coding model leaderboard ranks Claude Opus 5.5 first and GPT-6 Astra second, with Gemini 4 Argon down at ninth, which the poster calls surprising.

The poster says Gemini 4 Carbon is being tested on Arena under the name Gemini-4-Argon, may ship within weeks, and claims his own tests beat Argon.