Just so you know where my opinion is coming from, I've been designing and teaching materials design using frequency-based vocabulary since 2008 and have given Nation's and similar levels tests to hundreds of students in several countries from true beginners to PhDs teaching EFL. I helped devise the bilingual Gujarati test that is on Nation's website and set up the levels testing for 12 colleges at University of Delhi. I've published 8 textbooks (most available as free download, if anyone wants to try them) using frequency-based tools. All that is to say, my opinion on this issue is descriptive, rather than prescriptive, based on what I've seen in my own work.
The goal in my testing has been to help learners become independent users who can gain new knowledge in English in an academic environment. In passive vocabulary testing across levels and cultures, I've had students consistently test much lower than prescriptive estimates while still functioning at adequate CEFR levels for their purposes.
In ESL classes, I've rarely had an PhD candidate in the U.S. test much above 3k words and, yet, they function, largely because they work in a sphere where they also know the technical/low-frequency vocabulary. For professors of English abroad, a few close to C2, many B2, others as low as B1, I never had more than one or two reach 10k. The vast majority were 3-5k if they were functioning as teachers. In some cases, teachers were below 2k (below B2), but they needed support at that level.
We still don't know exact numbers for each level, but, my estimates are these: Based on tests, it only takes about 600 headwords to carry on a simple conversation of the type IEPs often use for entrance interviews. I'd put those learners at A1. Students cannot read sufficiently at all without the first 1000, so this is the first goal. Once they are solidly there, they are generally low-mid A2 in other skills. A2 is a fairly quick phase. After that, it gets fuzzy, but they cannot reach B2 (independent user) without passive comprehension of the first 2000 (GSL). After that, we jump to focusing on the AWL. Beyond that, it really doesn't matter for most learners because, at B2, they can function in a mainstream academic setting reading authentic texts to learn new information in English and gain vocabulary incidentally. They will likely still need support in writing, but they probably don't need special materials or IEP classes for most purposes.
Between those two levels (low A2 - B2 independent), we only need to sort students broadly. It's ok for classes to produce at different levels. We just need them to all understand a similar level of input, i.e. the teacher and books. We see a natural break in classes around 1200, again at 1500, then 1800 and on to 2000+ the first half of the AWL, taking them from low A2 to high B1 and on to low B2. B1 seems to begin around 1500 and is a long, slow, broad level. It can easily take twice as long as the move through A2. Once they're into the AWL, they are generally low B2, but they can't be B2 without the 2k. I believe this is because they simply can't understand enough in context without the 2k to learn new words from context.