> So what is happening is that models are broadening difficult-to-reason-to skills on benchmarks. ... If I'm being impolite I'd say the frontier models are trying to make sure they'll pass the "gotcha" interview questions for many jobs.
These "gotcha" skills will be needed for frontier research/scholarly tasks. It makes sense to focus on these now that traditional domains like coding or frontend design are broadly saturated (i.e. Astra- or Fable-level outcomes) even by lower-tier open models.
Real "gotcha" skills, yes (perhaps, they're gotchas because the underlying theory is not well understood. More can be achieved, certainly for superintelligence, by forcing understanding of the underlying theory rather than the gotcha). AND the incentives are totally misaligned.
"Gotcha skills", if we're going to call it that, are so popular because the AI labs are buying "tasks" to train on at huge prices, but mostly if they're gotchas.
The problem with that is that a collection of facts in the positivist sciences (and math) are always incomplete and almost preclude a full understanding of math, because it takes away the reason (to an extent) for full understanding.
What has happened to artificial analysis is that they've begin to focus a lot on very broad tasks like legal reasoning, using very difficult to work with scientific software packages, correctly analyzing biology experiments (without or only tiny amounts of explanation, count cells, BUT leave it to the model to figure that cell that aren't round, or have a distorted cell membrane (ie. have burst) are dead), very difficult physics reasoning ("2 electrical wires are close together with power flowing in the same direction. What happens?", and you get points deducted if you don't realize the problem with high voltage AC power, where the wires will shake themselves loose if the phase doesn't match perfectly, meaning the wires need to be EXACTLY the same length or ...). If I'm being impolite I'd say the frontier models are trying to make sure they'll pass the "gotcha" interview questions for many jobs.
The 3 differences:
Qwen 4 is a huge jump from Qwen 3.8 in coding
Qwen 4 is a big jump in using it for agentic workflows
Qwen 4 also performs much better on avoiding hallucination
If the difference between Qwen 4 Max and Qwen 4 flash is similar to 3.8 and 3.8 flash next (ie. about 10%), then 3.8 flash next may just barely beat Opus 5.4 at coding.