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

A user lists Opus 5.5, Astra 6.1, Muse Wonder Full and Gemini 4 Argon and asks which is the best AI model right now.

A developer says a few Argon thinking, research and planning sessions a week justify the $20/month Gemini subscription, paired with a local Qwen model and Claude Opus 5.5.

A user had Gemini 4 Argon (high) paint an original Friedrich of its choosing, with more paintings and process at stillwet.art.

Gemini 4 Argon places 6th overall on Design Creativity Bench with a creativity score of 55.7, ahead of Claude Opus 5.5 and Fable 5.1, and 98.6% design requirement adherence, up 1.1% from Gemini 3.8 Flash.

A roundup of developer complaints, including from Theo, says Argon hallucinates about how systems work and was factually wrong on over 50% of claims in tests, making it unsuitable for coding.

User HBCoop_ posts a first video of results from Gemini 4 Argon using a quick prompt about filmmaking use cases.

Anthropic's Opus 5.5, Sonnet 5.5 and Fable 5.1 hold the top three spots on the AA Intelligence Index, with GPT-6 Astra 4th and Gemini 4 Argon 5th.

The author used Blender renders as a base, Nano Banana to improve the visuals and Omni Flash for smoother motion, with Argon coordinating the work in about one third the time GPT-6 took on similar videos.

The same prompts were run on architecture, character animation, photoreal scenes, game menus, a Rocket League clone and a Minecraft prototype, finding Gemini cheaper and faster while Opus wins on quality for longer builds.

Mollick says Google has long been good at finding new uses for its models, like deep research and NotebookLM, but saw no similar product innovation at the Gemini 4 launch, which felt more like following trends.

A user suggests Google put the same Gemini 4 Argon usage limits into Antigravity Pro instead of the Gemini app, so developers can use it for coding and see the Pro plan's value.

The live New York show features the Google DeepMind team discussing Gemini 4 Argon, alongside Meta Muse Wonder and Anthropic's numbers.