I've already been using v4-flash extensively. It's genuinely productive at a high quality level, and the price is very appealing.
The instruction-following capability is excellent. I assign multiple tasks at once, and it iterates meticulously until completion. Manual verification also finds no issues. Additionally, the token speed is fast, which speeds up the actual task completion time. I'm using opencode + v4-flash.
I suspect instruction-following might be two sides of the same coin as long-context capability — many models tend to forget earlier instructions as they go along. As for the pro version, I still find it a bit expensive. I haven't used it many times, and my spending is roughly the same as with flash. I'm waiting for an official price drop in the second half of the year when hardware supply is sufficient, or for third‑party providers with ample hardware (yes, Tencent, I'm looking at you) to adopt it. Based on the FLops estimates in the paper, a price of pro being roughly three times that of flash seems reasonable.