DeepSeek Opens Inference Framework Amid Ban

Chinese startup DeepSeek open-sourced DeepInfer, an inference framework for its V4.1-Flash model, a day after the Trump administration banned federal civilian agencies from using the tool. The system costs $38 per GPU per hour on Nvidia and Huawei chips, undercutting OpenAI's GPT-6 Astra pricing of $10 and $50 per million tokens.