Reddit discusses DeepSeek v4 Flash cost efficiency

DeepSeek v4 Flash has 284B parameters but is priced very low on providers like OpenRouter for inference. Users speculate on how this is possible, with some citing quantization and efficient architecture.