Oddly, the Chinese LLM host SiliconFlow only makes it available with 32k context, which is even smaller than their DeepSeek-R1 offering.