GLM 5.3 vs Qwen 3.8 Flash
Compare models
Qwen 3.8 Flash is more worth it — 74% vs 58%
How we score worth-it? →Solid, unremarkable
- It seems that GLM-5.3 is still the go-to for cybersecurity.— hn
- 5.3 is much stronger — it barely errors anymore, and intent understanding is much better than before.— v2ex ↗
- Personally I find GLM-5.3 to be smart enough to be my daily driver, writing code, doing investigations, etc., including multi-codebase changes and finding the relevant documentation/supporting code for itself so I don't have to explain everything. It's fast, cheap, and smart enough... I do wish the bigger model could use vision though — that's currently limited to the flash model.— hn ↗
- GLM 5.3 produced this after a bit of prodding.— hn ↗
- Right now, I'm using Opus 5.5 as the main model, Longcat 2.5 for all subagents, and GLM 5.3 for adversarial security PR reviews.— hn ↗
- I subscribed to Pro, and I'm in an 'avoid using it unless I have to' state — I only turn to it when Codex, Grok, and DeepSeek are all unusable, because it's really, really slow.— v2ex ↗
- A problem I'd been grinding on with GLM5.3 for weeks was solved perfectly by fable5.2 in 3 hours for 200 yuan... As a long-time GLM Max user, I was stunned — the first time I've felt such a huge capability gap.— v2ex ↗
- This is a summary of what I encountered with GLM such that I had to implement the correct solution myself since it took too long for an otherwise trivial refactor.— huggingface ↗
- 要不像 GLM 这种的,在模型能力不行的情况下(最近降智特别严重),同时还慢、贵— v2ex ↗
- I haven’t been impressed with this model and I wanted to be. This is a case where benchmarks just don’t translate.— youtube ↗
“Solid and dependable. Does what it says on the box.”
“Unremarkable but reliable — which is exactly what some teams want.”
“Lacks a standout killer feature.”
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.”
Which should you pick?
The gap is clear: Qwen 3.8 Flash sits at 74% while GLM 5.3 is at 58%, a 16-point difference. If you just want the safer pick, start with Qwen 3.8 Flash.
On entry price, Qwen 3.8 Flash is cheaper ($0.11 vs $1.4, GLM 5.3 costs about 1173% 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 GLM 5.3 or Qwen 3.8 Flash more worth it?
Qwen 3.8 Flash is more worth it, with a 74% worth-it score vs 58% for GLM 5.3.
How much does GLM 5.3 cost vs Qwen 3.8 Flash?
GLM 5.3: $1.4/$4.4 per 1M · Qwen 3.8 Flash: $0.11 / $0.38 per 1M tokens.
What do users say about GLM 5.3 vs Qwen 3.8 Flash?
GLM 5.3: Solid and dependable. Does what it says on the box.. Qwen 3.8 Flash: Sparse and fast — only 6B active parameters, decoding at 150 tps on a single consumer GPU..