← 全部模型

Qwen · 近 7 天

Qwen3.8 27B

0–100 分,越高代表社区使用体验越正面,不代表能力测试成绩。

更新于 2026-09-05 14:21 UTC · 2026-08-29 – 2026-09-05 UTC
社区口碑分66.8/100360 条评论 · 近 7 天

社区评论

精选近 7 天的不同意见,摘录条数不代表真实好评比例。

24 条精选摘录

正面通用文本

“到目前为止使用3.8版本,我只需要检查最终结果...”

Redditen机器翻译
展开原文

So far with 3.8 i just need to check the end result...

正面本地部署

“q3 xl unsloth。150k上下文。hatch 512和kv q4。表现稳定。审查luna和sol计划,后续确认了发现的问题。”

Redditen机器翻译
展开原文

q3 xl unsloth. 150k context. hatch 512 and kv q4. Its solid. review luna and sol plans and catch gaps confirmed later by them

正面编程

“它能够遵循我的编码规范,甚至能够正确地记录架构而不需要太多指导。没有其他能够装入我的16GB显存的模型能做到这一点。”

Redditen机器翻译
展开原文

it can follow my coding conventions and even document the architecture the right way without much hand holding. No other model that fit on my 16GB VRAM does as well.

正面通用文本

“Qwen3.8 27b是一个非常强大的模型,尽管追求更大的模型需要大量投入,但收益甚微。”

Redditen机器翻译
展开原文

Qwen3.8 27b is a very capable model, and there isn't much to gain for pursuing larger models despite the significant investment required.

正面速度与延迟

“4090 256k 速度是45tok/s,降低到128k估计有55左右”

负面编程

“Qwen3.8 27B 的行为与 Minimax M3 非常相似(思考很多,检查一切,并且喜欢在长时间会话中独立继续工作直到完成)。它就像是略微弱一点的 M3。”

Redditen机器翻译
展开原文

Qwen3.8 27B have very similar behavior to Minimax M3 (thinks a lot, check everything, and like to continue working on its own for long session until it's done). It's like a slightly worse M3.

正面推理

“Qwen 3.8 27B 确实名不虚传,但请记得使用 froggeric 模板和/或中等推理”

Hacker Newsen机器翻译
展开原文

Qwen 3.8 27B is the real deal BUT remember to use froggeric template and/or medium reasoning

正面速度与延迟

“Qwen 3.8 27B 在 KV 缓存方面效率极高”

Redditen机器翻译
展开原文

Qwen 3.8 27B Is extremely efficient when It comes to KV cache

正面本地部署

“它们运行得非常好,在 16GB 的 5060 上运行得非常好”

Redditen机器翻译
展开原文

they work very well, like very very well in a 5060 with 16GB

正面本地部署

“开箱即用,与 Pi 配合得非常好”

Redditen机器翻译
展开原文

works very well with Pi out-of-the-box

褒贬兼有速度与延迟

“它思考个没完,有时候还会陷入循环”

Redditen机器翻译
展开原文

it thinks forever and sometimes in a loop

负面推理

“它会在历史事实上产生幻觉,所以不要过于信任”

Redditen机器翻译
展开原文

It can hallucinate with historical facts so dojlnt trust too much

负面速度与延迟

“qwen3.8 27b 工作流程相同但速度慢得多”

Redditen机器翻译
展开原文

qwen3.8 27b is the same workflow but sloooooweeeeeerrrrrrr

负面推理

“它曾在一个关键细节上产生过幻觉”

Redditen机器翻译
展开原文

it had hallucinated a key detail

负面通用文本

“感觉它永远搞不懂如何更新记忆”

Redditen机器翻译
展开原文

It feels like it can never figure out how to update memory

负面速度与延迟

“上次使用时完全无法使用。用了不到几分钟就遇到速率限制。它很快但无法维持其声称的速度,立即就会达到速率限制。如果你无法完成...”

Hacker Newsen机器翻译
展开原文

Completely unusable last time I tried it. You got hit with rate limits after the first few minutes of using it. It's fast but cannot sustain its claimed speed before it immediately hits its rate limit. What good is it if you can't finish a

负面速度与延迟

“它思考了15分钟,编码了2分钟”

Redditen机器翻译
展开原文

il a réfléchi 15 min et codé 2 min

负面通用文本

“在qwen网站上没有3.8-27B型号”

Redditen机器翻译
展开原文

in qwen website nothing models 3.8-27B

中性通用文本

“说实话,它并不完美,但如果你能给出直接的指令,它还是能用的”

Redditen机器翻译
展开原文

It's not perfect, not going to lie, however it's usable if you can give direct instructions

负面本地部署

“我看到内存使用在工具使用过程中不断增长,使用qwen3.8 27B时这肯定有一定影响”

Redditen机器翻译
展开原文

i see the memory use growing during tool use, with qwen3.8 27B it certainly plays a role

负面推理

“在多步骤代理任务上会崩溃。试过用它做一个需要跨文件搜索和编辑的重构,到了第三次工具调用时它就已经在引用那些它已经覆盖掉的函数”

Redditen机器翻译
展开原文

falls apart on multi-step agentic tasks. Tried it on a refactor that needed search plus edit across files and by the third tool call it was citing functions it had already overwritten.

负面编程

“temp 1无论用什么量化都会很烂,0.7才是该用的temp,不知道他们为什么推荐temp 1”

Redditen机器翻译
展开原文

temp 1 will suck no matter the quant, 0.7 is the temp to use, idk why they reccomended temp 1

负面本地部署

“性价比 远不如 api”

负面本地部署

“干活的话稍微复杂一样就全崩,我现在本地的只能纯娱乐想要干活还是得拉云端模型”