Everyday answers
GPT-6.1 Sol gives polished, well-structured answers and is good at reading between the lines of a vague request. Gemini 3.8 Flash is quicker and more literal. For drafting and explaining, Sol usually needs fewer follow-ups.
edge: GPT-6
Cost per message
Gemini 3.8 Flash costs about 6 tokens for a typical reply on LazySusan, against 16 tokens for GPT-6.1 Sol. At the top, Gemini 3.1 Pro is about 18 tokens and GPT-6 Astra 77 tokens. GPT-6 Luna is the cheapest of the lot at about 1 token.
edge: Gemini
Big and mixed inputs
Gemini takes in long PDFs, images and spreadsheets comfortably and accepts prompts up to 1M tokens on LazySusan. GPT-6 models accept up to 400K, which is plenty for most work but not for everything.
edge: Gemini
Hard problems
GPT-6 Astra is OpenAI's model for the hardest reasoning, maths and code. Gemini 3.1 Pro is strong and cheaper per reply. If you only escalate occasionally, Astra is the safer bet; if you escalate often, 3.1 Pro's cost adds up more slowly.
edge: GPT-6
Images
Each company's image model is a separate tool on LazySusan: GPT Image for OpenAI and Nano Banana for Google. Nano Banana is the stronger editor of existing photos; GPT Image follows long, detailed instructions very closely.
edge: Tie