Qwen 3.8 Flash vs Qwen4-Max
Comparer des modèles
Qwen 3.8 Flash is more worth it — 74% vs 39%
How we score worth-it? →Alibaba's fast, cheap open-weight multimodal model (Qwen 3.8 family) with 1M context, deep thinking, and agentic coding — the budget Flash sibling to Qwen 3.8 Max.
- Qwen 3.8 Flash Next has only 6B activated parameters — on a single PRO 6000 it pre-fills over 10K tokens and decodes at 150 tps, with room to optimize. It really is sparse and fast.— v2ex ↗
- Qwen 3.8 flash next is way better than 27B. It's so good I don't even use Claude anymore.— hn
- Qwen Flash Next not only seemed less delusional than the 27B, but also made useful observations and problem-solved ways around many website access issues.— hn ↗
- Qwen3.8-next-flash is impressive though, especially with Strata or Basalt as inference engines. On a 5090 laptop with 64GB RAM one can run NVFP4 quants at 120-140t/s decode and 3000-4000t/s prefill.— hn ↗
- Qwen 3.8's Flash Next at A6B seems about juuuuust right— hn ↗
“Sparse and fast — only 6B active parameters, decoding at 150 tps on a single consumer GPU.”
“Flattened the VRAM barrier — a data-center-grade long-context model that runs on a consumer RTX 4090.”
“Superior quality for coding, and notably good at finding issues in generated code.”
Alibaba's 2.4T-parameter MoE flagship (71B active), native vision, 1M-token context — 6th on Code Arena.
- This is exciting. The continued focus on efficiency alongside quality is encouraging — it makes me hopeful to see what smaller models will become capable of.— hn ↗
- Judging from just these few tests, it's the best among the domestic models.— bilibili ↗
- Coding is pretty bad. On FrontierCode, Qwen 3.8 scores 23.4%, Qwen4 26.5%, DeepSeek Flash 28.5%.— v2ex ↗
- In our short horizon data analysts exam, it scored a few points worse than Qwen 3.8 27B from August (which is open weights) and cost a bit more. There's about 10 models that we've tested that have scored both better and cheaper.— hn ↗
- So qoduck was Qwen4 Max Preview all along? I thought it was qwen4-27b — I didn't feel any 'Max' capability in it.— v2ex ↗
- I knew there wouldn't be any improvement, but I didn't expect a regression. The Qwen team is decent at foundation models but really has no idea how to do post-training.— bilibili ↗
- I tested it — it's pure garbage. It can't understand a handover doc, it's even lazier than Feiyu, and whatever it says from its seemingly big knowledge base needs verification and turns out to be all nonsense. It never examines its own decisions or code while executing. The 300+ yuan was just a donation to Alibaba; I don't owe it anything.— bilibili ↗
“2.4T total parameters with only 71B active per token — lower inference cost at the same scale.”
“Native vision plus a one-million-token context window for long-horizon work.”
“Ranked 6th on Code Arena, with strong agentic and visual-reasoning benchmark scores.”
Which should you pick?
The gap is clear: Qwen 3.8 Flash sits at 74% while Qwen4-Max is at 39%, a 35-point difference. If you just want the safer pick, start with Qwen 3.8 Flash.
On entry price, Qwen4-Max is cheaper ($0.075 vs $0.11, Qwen 3.8 Flash costs about 47% more).
Both are in the Chat category with similar coverage; the real difference is stability and cost — run one of your own tasks through each before deciding.
What users actually said about Qwen 3.8 Flash: "Qwen 3.8 Flash Next has only 6B activated parameters — on a single PRO 6000 it pre-fills o…"
FAQ
Is Qwen 3.8 Flash or Qwen4-Max more worth it?
Qwen 3.8 Flash is more worth it, with a 74% worth-it score vs 39% for Qwen4-Max.
How much does Qwen 3.8 Flash cost vs Qwen4-Max?
Qwen 3.8 Flash: $0.11 / $0.38 per 1M tokens · Qwen4-Max: $0.075/M input (cache hit).
What do users say about Qwen 3.8 Flash vs Qwen4-Max?
Qwen 3.8 Flash: Sparse and fast — only 6B active parameters, decoding at 150 tps on a single consumer GPU.. Qwen4-Max: 2.4T total parameters with only 71B active per token — lower inference cost at the same scale..