Two companies. Two announcements. Two days.
On March 16, Mistral released Mistral Small 4, a multimodal model built on Mixture of Experts architecture. The next day, OpenAI released GPT-5.4 mini and GPT-5.4 nano for cost-sensitive applications. Both companies targeted the same buyer: teams that need serious AI capability without flagship pricing. Neither set of benchmark claims has been independently verified by Epoch AI or a third party at the time of publication. Both sets are vendor-reported. That context belongs at the front, not the footnotes.