Image
DeepSeek V4 Pro vs Gemini 3.6 Flash: cheaper, but 3–6× slower to a finished answer
Last month's story was the DeepSeek V4 Pro price cut. This week's AI Arena run is the benchmark follow-up — and it handed us a number we refuse to publish as fact: DeepSeek V4 Pro reported 4,096,000 tokens per second. That is physically impossible, so we pulled the raw JSON and recomputed it ourselves.