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

The post claims Gemini 4 Carbon's early access program has begun and the model is being tested on LMArena, billed as Google's Opus killer.

A leak says Gemini 4 Carbon early access is open to partners and testing on LMArena, with clearly better 3D than Argon but still poor frontend/UI and slower speed.

A personal October 2026 AI tool ranking lists Gemini 4 Argon as the best multimodal understanding model, with Claude Opus 5.5 and GPT-6.1 Sol topping coding.

The post claims Gemini 4 Argon costs two to three times less than other frontier models while matching top-tier performance on robotics learning engineering.

Google's Koray says Gemini 4 Argon processed nearly 25 trillion tokens in its first 72 hours, three times what Gemini 3 did.

Google's Logan Kilpatrick says Gemini 4 Argon processed roughly 25 trillion tokens in its first 72 hours, calling demand still underappreciated.

A user reports Gemini 4 Argon feels more natural than Opus 5.5 and holds long-context agent tasks well, but codes worse than GPT and Claude and burns quota fast.

A post flags Gemini 4's 78% on FrontierMath Tier 4 as far below OpenAI models that already saturate the benchmark, and asks where Google's Deep Think is.

A short image post captioned as the Gemini 4 Argon experience, with no details on what was actually tested.

A roundup of seven upcoming model leaks says Google is testing a Gemini 4 Carbon checkpoint on Jetski, with staff comparing its coding feel to Opus 5.5, though it is unclear whether it is an Argon update or a separate release.

A poll asks which model is best for coding among GPT-6.1 Sol, Claude Opus 5.5, Gemini 4 Argon and Grok 4.7.

A week after Google released Gemini 4 Argon, users are building 3D worlds, interactive machines, physics simulations and games; the post collects 10 examples.