[00:02] for the smarter model. Another picked the cheaper one and shipped [music] the same week. Which one made the right call? Meet Priya and Marco, two startup founders. On September 3, OpenAI released GPT-6 [music] [00:14] Astra, its flagship and Priya bought it. Then on September 22 came GPT-6 Soul. >> [music] >> just below Astra and above the fast Luna >> just below Astra and above the fast Luna tier and it costs $2 per million input [00:28] [music] tokens so in $10 per million output tokens. Astra's price is 10 and [music] $50. That's five times cheaper per token and Marco picked it. Did that number surprise you? [music] That's the power of a price tag. Now watch what [00:41] happened. OpenAI says [snorts] Soul [music] beats Claude Opus 5 on a business workflow test called automation bench at 9% of the cost per task. [music] Marco's bill dropped and his work still shipped. But before you copy [00:54] him, here's the [music] honest part. Soul's best score is actually lower than the previous GPT-5. 6 Soul's best on [music] Deep Sweswi 68.8% [music] Deep Sweswi 68.8% against 72.7% [music] and on OSWorld 2.0 [01:08] 64.4% against 66.2%. Soul wins when the budget is equal, [music] not when you need the absolute ceiling. And OpenAI's half-price claim compares against a promotional rate so the real saving is smaller. [music] [01:22] So here's a simple rule. Use Soul for everyday coding, workflow automation, and [music] bulk tasks. Keep the flagship for the one hard problem that really needs it. [music] Priya or Marco, which one are you? Comment below. [01:35] Subscribe for more such tech insights. [music]