Of the major artificial intelligence models, DeepSeek's new V4-Flash is the cheapest to run, according to a new study from research firm Artificial Analysis. The firm compared the token prices it costs leading models to run benchmark tests, with DeepSeek's averaging 3 cents per test. Meanwhile, fellow Chinese company Moonshot AI's buzzy Kimi K3 model cost 86 cents per test. As for U.S. companies, OpenAI's GPT-5.6 Sol cost $1.86, while Anthropic's Claude Fable 5 cost $3.15. s...