Gemini 4 Argon - Hackdot

Re:Gemini 4 Argon by robertbarbe on Thursday October 01, 2026 @03:20AM (#49918725)

Google's way of presenting their model's performance is terrible compared to Anthropic or Openai. Google's benchmark results are not useful at all. If you look at their methodology, they only present the model's performance at the Max reasoning levels. https://storage.googleapis.com/deepmind-media/gemini/gemini_...

I don't know anyone who normally uses those models at these levels where Work output is excruciatingly slow.

[ Parent | on HN ]

Re:Gemini 4 Argon by ostwilkens on Thursday October 01, 2026 @03:22AM (#49918741)

I almost exclusively use Max reasoning for all my work

[ Parent | on HN ]

Re:Gemini 4 Argon by alpineman on Thursday October 01, 2026 @05:01AM (#49919408)

I used to do that too but found that often, unless it's a really tough/unclear bug, lower reasoning levels are much better and don't overcomplicate the code. It also writes clearer explanations and plans on lower thinking levels.

[ Parent | on HN ]