OpenCraft Launch Reports

What every model actually makes

When a new model ships, we run it through the same frozen prompts — drawings, a deck, a playable game — on the real product, and publish every output next to the current gold standard, with the cost and time of each. No benchmark scores; the files themselves.

001GPT-5.6 Luna vs DeepSeek V4 Flash Vision ExpTwo models, the same frozen prompts, run through the real product. Every drawing here is the model's own SVG as delivered, with what each one cost and how long it took.openai/gpt-5.6-luna · deepseek/deepseek-v4-flash-vision-exp