I needed an LLM API for a small side project — a tool that turns messy meeting notes into a clean summary — and landed on Qwen mostly because a few people recommended it for long-document handling. What I didn't expect was how much time I'd spend just figuring out which tier to use, since Qwen ships flash/plus/max variants and the docs don't exactly hold your hand on picking one.
So I did the boring-but-useful thing: ran the same 10 real meeting transcripts through all three tiers and actually read the outputs side by side, instead of guessing based on the names.
What I found, for my specific task:
flash missed some nuance in longer transcripts — noticeable but not disqualifying for a quick first draft
plus was the sweet spot — caught the nuance flash missed, cost meaningfully less than max
max was marginally better on a couple of trickier transcripts, but not enough to justify the cost difference for my use case
I access all three through RouteAI, so switching tiers to test was a one-line change instead of three separate integrations — genuinely doubt I would've bothered running this comparison otherwise, given how side-project time works.
Ended up on plus for production. Your task might land differently — the point isn't "use plus," it's that the tier names don't tell you enough on their own, and a same-day comparison on your actual data is worth the hour it takes.
TL;DR: Tested Qwen's flash/plus/max tiers on the same real task instead of guessing from the tier names. plus was the best cost/quality balance for my specific summarization use case — but the real point is to test your own data, not copy my conclusion.
Top comments (0)