Qwen 3.8 27B is excellent, but it defaults to overthinking things
Qwen 3.8 27B is excellent, but it defaults to overthinking things
Qwen 3.8 27B 表现出色，但默认设置下容易“过度思考”
Friday’s big release was Qwen 3.8 27B, an Apache 2 licensed 27B parameter vision-capable LLM from Alibaba’s Qwen research lab. I’ve been looking forward to this one: 27B is an excellent size for running a model on a reasonably specced laptop, and its predecessor Qwen 3.6 27B was impressive.
周五发布的大新闻是 Qwen 3.8 27B，这是阿里巴巴 Qwen 研究实验室推出的一款采用 Apache 2 协议开源、具备视觉能力的 270 亿参数大语言模型。我一直很期待这个版本：27B 是在配置尚可的笔记本电脑上运行模型的绝佳尺寸，而它的前身 Qwen 3.6 27B 就已经令人印象深刻。
Qwen’s self-reported benchmarks for this model are eye-opening. They show a boost from both Qwen 3.6 27B and the closed-weight Qwen 3.7-Plus, which was one of Qwen’s strongest models of any size as recently as May this year. It will be interesting to hear what independent benchmarks have to say about the model.
Qwen 官方公布的该模型基准测试结果令人眼前一亮。数据显示，它不仅超越了 Qwen 3.6 27B，还胜过了闭源的 Qwen 3.7-Plus——后者在今年 5 月时还是 Qwen 各尺寸模型中最强的一款。后续独立基准测试对该模型的评价将会非常值得关注。
I’ve been running the model on two different machines: my 128GB M5 Max MacBook Pro, and an NVIDIA DGX Spark. On both machines I’m running LM Studio and their 17GB Q4_K_M quantized build. I also tried using llama-server directly on the Spark.
我已经在两台不同的机器上运行了该模型：一台是配备 128GB 内存的 M5 Max MacBook Pro，另一台是 NVIDIA DGX Spark。在这两台机器上，我使用的都是 LM Studio 及其 17GB 的 Q4_K_M 量化版本。此外，我也尝试在 Spark 上直接使用 llama-server。
The default of extra high results in spectacular over-thinking
“Extra high”默认设置导致了惊人的过度思考
Qwen’s documentation describes the model as defaulting to xhigh for the reasoning effort, and the LM Studio GGUF I’ve been trying preserves that default: Qwen3.8 comes with official support for reasoning_effort, which can be used to adjust reasoning depth and control cost:
xhigh(default): for complex tasks demanding thorough analysismedium: balancing accuracy and speedlow: efficient reasoning optimizing for speed and cost
Qwen 的文档显示，该模型在推理强度（reasoning_effort）上默认设置为 xhigh，而我尝试的 LM Studio GGUF 版本也保留了这一默认值。Qwen 3.8 官方支持 reasoning_effort 参数，可用于调整推理深度并控制成本：
xhigh（默认）：适用于需要深入分析的复杂任务medium：平衡准确性和速度low：针对速度和成本优化的高效推理
This is a hilarious default. It’s absolutely not a good way to run the model, especially on consumer hardware. I’ve been finding the results extremely entertaining. I quickly ran into problems with LM Studio’s default context limit of 8,192 tokens—Qwen was using them all up thinking about even the most mundane of problems. I loaded the model with the full 262,144 maximum context length and that problem went away.
这是一个滑稽的默认设置。这绝对不是运行该模型的最佳方式，尤其是在消费级硬件上。我发现其结果非常有趣。我很快就遇到了 LM Studio 默认 8,192 token 上下文限制的问题——即使是处理最平凡的问题，Qwen 也会耗尽所有 token 来进行思考。当我将模型加载到 262,144 的完整最大上下文长度时，这个问题就解决了。
Here’s the pelican riding a bicycle SVG I got from my first attempt with that increased context length. It took 21 minutes to generate, using 22,276 reasoning tokens to produce 3,223 tokens of output. You can read the reasoning trace here.
这是我在增加上下文长度后，第一次尝试生成的“鹈鹕骑自行车” SVG 图片。它耗时 21 分钟，使用了 22,276 个推理 token 才生成了 3,223 个输出 token。你可以在这里阅读它的推理过程。
Was that worth waiting 21 minutes for? Absolutely not. Here’s the same prompt run with reasoning turned off—transcript here. This one produced 3,715 tokens and took 137s—just over two minutes.
为了这个结果等待 21 分钟值得吗？绝对不值。这是关闭推理功能后运行相同提示词的结果——记录在此。这次生成了 3,715 个 token，仅耗时 137 秒，也就是两分多钟。
I said Qwen at xhigh has a tendency to over-think things, but how bad really is it? I tried a much simpler prompt, again with that default extra high setting: draw an svg of a circle. Qwen’s reasoning trace started like this: “The user is asking for an SVG drawing of a circle. Simple request — but I want it to be a carefully crafted piece. Let me make something that goes beyond just
我说 Qwen 在 xhigh 设置下有过度思考的倾向，但到底有多严重呢？我尝试了一个更简单的提示词，依然使用默认的超高设置：draw an svg of a circle（画一个圆的 SVG）。Qwen 的推理过程是这样开始的：“用户要求画一个圆的 SVG。这是一个简单的请求，但我希望它是一件精心制作的作品。让我做一些超越简单
Several minutes later it produced this absolutely beautiful animated circle, which was entirely not what I had asked for!
几分钟后，它生成了一个绝对精美的动画圆，但这完全不是我所要求的！
My strong recommendation: ignore that default. Run Qwen 3.8 27B on low or even no reasoning levels at first. It’s a great model, but wow that default setting is a bad place to start.
我的强烈建议是：忽略那个默认设置。起初运行 Qwen 3.8 27B 时，请使用 low 甚至关闭推理级别。这是一个很棒的模型，但说实话，那个默认设置绝对不是一个好的起点。
It’s very good at bounding boxes
它在边界框识别方面非常出色
A fun way to test a vision model is to see how well it can return bounding boxes around items in a photograph. I’ve seen previous Qwen models deal well with this, so I decided to put it to the test drawing bounding boxes around some pelicans.
测试视觉模型的一个有趣方法是看它能否准确地为照片中的物体绘制边界框。我见过之前的 Qwen 模型在这方面表现良好，所以我决定测试一下它为鹈鹕绘制边界框的能力。
Building a tool to label bounding boxes
构建一个标注边界框的工具
That visualization of the bounding boxes was taken using a new custom tool that I had Qwen 3.8 27B build for me, running offline on my laptop. I forgot to dial down the thinking effort so it was massively over-engineered, but it did manage to produce this full interface from this single prompt: “Build an HTML page which has an input box for accepting the URL to an image and a textarea for accepting…”
上述边界框的可视化效果，是我让 Qwen 3.8 27B 在我的笔记本电脑上离线构建的一个新自定义工具生成的。我当时忘了调低推理强度，所以它过度设计得非常厉害，但它确实仅凭这一个提示词就成功生成了完整的界面：“构建一个 HTML 页面，包含一个用于输入图片 URL 的输入框，以及一个用于接收……”