benchmarks

2 posts tagged with "benchmarks"

Prompts Outlast Models: Testing GPT Image 2 Low

April 23, 2026

I tested gpt-image-2 low against the prompt corpus from my last two posts. 88 images, $1.15, zero rewrites. Prompts outlast models when the grammar is structural.

image generationgpt-image-2prompt engineeringbenchmarks

Read article

Trading Cents for Pixels: An Image Model Bake-off

April 10, 2026

12 min read

I compared OpenAI, Imagen 4, and Nano Banana for editorial travel images. The first winner looked right, but it wasn't the most accurate.

image generationbenchmarks

Read article

Browse by tag

architecture benchmarks career & culture claude code gpt-image-2 image generation mcp open source planning prompt engineering