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

A user says Gemini 4 Argon has outstanding writing and rare literary talent, but its conversation feels disjointed, lacks a stable identity, and is a clear downgrade.

The author used Gemini 4 Argon to recreate a Pokémon Showdown-like Gen 9 battle, saying the builds and mechanics are fairly realistic.

Hermes Agent's major update spans 24,200 commits and adds one-click local model deployment, a 600-plugin catalog, and support for GPT-6, Claude 5.5 and Gemini 4.

The author ran the same prompt through Gemini 4 Argon and Opus 5.5, saying Argon did better than expected and that details are in the replies.

A user recaps the week: Meta released Muse Wonder, Google brought Gemini 4 Argon to Pro and Ultra, and Anthropic's Fable 5.5 showed up in Azure's catalog.

Google opened up access to Gemini 4 Argon, which leads the Vals Index for finance, coding, legal and tax tasks at introductory pricing of $2/$10 per million tokens.

Hermes Agent v0.22.0 adds Gemini 4 Argon support with up to 1M output tokens, lets desktop start without an API key, and adds style changes by chat and guest-identity connectors.

HIX AI says Gemini 4 Argon is now available in its multi-model workspace, alongside the other AI models it offers.

A Russian-language post claims Gemini 4 Argon solved a puzzle that had stumped humanity for 59 years in just a few hours, without naming the problem.

Google released Gemini 4 Argon and showed it teaching Apptronik's Apollo humanoid to play cornhole from a single human video, with the robot sinking seven shots in a row.

A user who tried Muse Wonder and Gemini 4 Argon says Argon is well-rounded and cheap while Muse feels more agent-focused, and both still trail Opus on hard tasks.

A developer used Gemini 4 Argon to build a small meeting app that lets you fog up your webcam lens with your breath and draw on it, with a demo video.