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

The poster claims Google is reportedly running RSI self-improvement on Gemini 4 Argon while OpenAI preps GPT-6 Astra 6.1, and wonders why Anthropic keeps shipping Sonnet/Opus/Fable 5.5 news.

Researchers can apply with an institutional email, lab homepage, ORCID or Google Scholar profile and a research description to get six months of AI Ultra including Gemini 4 Argon for free.

A heated post about Indian politics and caste attacks that ends by asserting Google's Argon is currently the best model in the world.

A user shares a directive written by Gemini 4 Argon, asks whether it seems reasonable, and says they were not having it.

After exhausting Codex and Claude quotas, a user ran Gemini 4 Argon for 12 hours, finding it great at code and finally tool use, good for systems programming and frontend, but too robotic for creative work.

A Japanese shop owner says they used Gemini's latest model Argon to create a video announcing the store's day off, showcasing the new model.

A Turkish user says their suspended X account was reinstated after Gemini 4 Argon wrote an appeal citing X's policies.

A post asks which model is better for motion graphics work, Opus 5.5 or Gemini 4 Argon.

Google DeepMind says four Gemini 4 Argon agents beat Minecraft without human help in 27 hours 14 minutes, costing $1,620 at Argon's planned 2027 price.

The post asks readers to test GPT-6.1 Astra, Claude Opus 5.5, DeepSeek V4.1, Gemini 4 and Grok on real work instead of benchmarks, and share which broke first.

On one animation task, Gemini 4 Argon finished in 8m33s for $0.68, GPT-6 Astra took ~15 minutes for $2.90, and Opus 5.5 nearly an hour for $8.37.

BudgetPixel compares DeepSeek V4.1 Flash and Gemini 4 Argon frame by frame on the same flying-car backflip clip to see which tells a clearer story.