DeepSeek V4.1 Flash Scores 81.2, 1.4% Cost of GPT-6 Astra

OpenDesign tested 13 AI models on design tasks, with OpenAI's GPT-6 Astra scoring the highest at 82.7 points. DeepSeek's V4.1 Flash achieved 81.2 points, completing tasks in 5.3 minutes at a cost of 0.023美元. GPT-6 Astra took 11.1 minutes and cost 1.61美元. Other models like Claude Fable 5.1 scored 80.3 points, taking 12.8 minutes and costing 3.66美元. DeepSeek's model utilizes 552 billion parameters but activates only 8 billion for prompts and 16 billion for responses, enhancing efficiency. This is not DeepSeek's first cost-effective model; its V4 Pro previously scored within 5% of Claude Fable 5 while being cheaper. OpenDesign's tests focus on reliable design output, not general reasoning. GPT-6 Astra, released on September 3, is noted for versatility but is slower and more expensive than DeepSeek's model. Delivery rates were 57.7% for V4.1 Flash, 60% for GPT-6 Astra, and 56.7% for Claude Fable 5.1.