OpenAI's GPT-6 Luna ($0.10 in, $0.50 out) undercuts deepseek-flash on list price, so DeepSeek is no longer the cheapest option.
DeepSeek is 2 to 84 times cheaper than GPT-6.1 Sol and GPT-6 Astra on a typical agent workload.
OpenAI reprices the whole request above 272K prompt tokens, while DeepSeek lists a single rate across its 1M window.
DeepSeek works with the OpenAI SDK by changing the base URL and model name, but reasoning controls and usage fields differ.
OpenAI documents US and EU data residency for some models; DeepSeek lists none, so regulated teams must review its policy.