{"slug": "qwen-3-8-27b-is-great-but-turn-off-its-brain-first", "title": "Qwen 3.8 27B Is Great — But Turn Off Its Brain First", "summary": "Alibaba's Qwen team released Qwen 3.8 27B on August 14, a 17GB model that scores 61.7 on SWE-bench Pro, outperforming Claude Opus 4.6 Max's 53.4, and runs offline on consumer hardware. However, its default `reasoning_effort` setting of `xhigh` causes excessive deliberation, as demonstrated by Simon Willison's MacBook Pro taking 21 minutes to generate 22,276 reasoning tokens for a simple SVG request; setting the parameter to `low` or `none` reduces the time to 137 seconds, a 9x speedup. The model is licensed under Apache 2.0 and hit #1 on Hacker News with 893 points, but community analysis suggests a possible scale misalignment in the `ssm_conv1d` layers that may require a retrained release from Alibaba.", "body_md": "Alibaba’s Qwen team released Qwen 3.8 27B on August 14 — a 17GB model that posts near Claude Opus 4.6 Max scores on software engineering benchmarks and runs fully offline on consumer hardware. [Simon Willison ran it on his MacBook Pro](https://simonwillison.net/2026/Aug/16/qwen-38-27b/) and asked it to draw a pelican on a bicycle. The model spent 21 minutes generating 22,276 reasoning tokens before producing the SVG. The same prompt, with one parameter changed, completed in 137 seconds. The model is not broken — but it ships configured like it is.\n\n## Why Developers Are Paying Attention\n\nQwen 3.8 27B is a legitimate milestone for local AI. At 27.78 billion parameters, it scores 61.7 on SWE-bench Pro — higher than Claude Opus 4.6 Max’s 53.4, according to Alibaba’s reported benchmarks. It accepts text, images, and video natively, ships with a 262,144-token context window, and is licensed under Apache 2.0, meaning free commercial use without restrictions or per-token fees. The quantized GGUF version fits in 17GB and runs on a 24GB GPU or a well-specced Mac with 32–48GB of unified memory. The day Alibaba pushed the weights, the model hit #1 on [Hacker News with 893 points](https://news.ycombinator.com/item?id=49324985). For the local AI crowd, this was a big deal.\n\n## The Problem: It Defaults to Thinking Too Hard\n\nQwen 3.8 ships with `reasoning_effort`\n\nset to `xhigh`\n\nby default. That setting is designed for “complex tasks demanding thorough analysis” — but the model applies it universally, including for trivial requests. Ask it to draw a simple circle in SVG and it will deliberate at length before producing an over-engineered animated result you didn’t want. Ask it to draw a pelican on a bicycle and Willison’s MacBook Pro spent 21 minutes producing 22,276 reasoning tokens to generate 3,223 output tokens. The output was fine. The wait was not.\n\nHacker News developers confirmed the same pattern. One commenter reported 17,576 reasoning tokens for a task where a competing model used 1,021. The community consensus: the model is technically excellent, but it needs configuration before it is practically useful.\n\n## The Fix: One Parameter\n\nSet `reasoning_effort`\n\nto `low`\n\n, `medium`\n\n, or `none`\n\n. That is the whole fix. With reasoning disabled, Willison’s pelican prompt went from 21 minutes to 137 seconds — about 9x faster. The model produced 3,715 tokens and the result was the same quality.\n\n```\nresponse = client.chat.completions.create(\n    model=\"qwen3.8-27b\",\n    messages=[{\"role\": \"user\", \"content\": \"your prompt here\"}],\n    extra_body={\"reasoning_effort\": \"low\"}  # none | low | medium | xhigh\n)\n```\n\nIf you are running the model via llama-server, pass `--reasoning-effort low`\n\nat startup. Unsloth Desktop has a thinking toggle in the UI. The default `xhigh`\n\nreasoning mode should be reserved for genuinely hard research and analysis tasks where you can afford the wait. For coding, writing, and most day-to-day use, start at `none`\n\nand escalate from there.\n\n## There May Be More to It Than Config\n\nA community researcher on HuggingFace went deeper. Analysis of the model weights identified possible scale misalignment in the `ssm_conv1d`\n\nlayers — specifically blocks 52–62 of the DeltaNet attention mechanism. The hypothesis: distorted temporal signals cause the model’s reasoning to not know when to stop, turning a configuration problem into a training artifact. A partial community fix already exists at [redashes/Qwen3.8-27B-BF16-SSMFIX](https://huggingface.co/redashes/Qwen3.8-27B-BF16-SSMFIX) on HuggingFace. This is not confirmed by Alibaba, and the `reasoning_effort`\n\nparameter remains the practical solution for most developers. But it suggests that a full fix may require Qwen to retrain and push corrected weights — possibly in a future release.\n\n## Local AI vs. Cloud: The Honest Math\n\nWith reasoning tuned down, Qwen 3.8 27B runs at 15–30 tokens per second on an M-series Mac. GPT-5.6 Luna through the API runs at around 184 tokens per second. That gap is not going away. Cloud APIs are faster and getting cheaper. For latency-sensitive applications, the comparison is not flattering.\n\nBut for the right workloads, the local value proposition holds. Zero per-token cost at scale. Complete data privacy — nothing leaves your machine. Offline operation for air-gapped or restricted environments. Apache 2.0 permanence, with no risk of pricing changes or model deprecation affecting your deployment. For batch processing jobs, coding agents that run overnight, or any application where a few extra seconds per request is acceptable, a properly configured Qwen 3.8 27B is genuinely competitive with hosted alternatives — and costs nothing to run after the hardware investment.\n\nThe model is good. The defaults are not. One parameter change separates a frustrating experience from a capable local AI that punches well above its weight class. Read [Willison’s full analysis](https://simonwillison.net/2026/Aug/16/qwen-38-27b/) and check the [official model page on HuggingFace](https://huggingface.co/Qwen/Qwen3.8-27B) before you write it off based on first impressions.", "url": "https://wpnews.pro/news/qwen-3-8-27b-is-great-but-turn-off-its-brain-first", "canonical_source": "https://byteiota.com/qwen-3-8-27b-overthinking-fix/", "published_at": "2026-08-17 09:07:58+00:00", "updated_at": "2026-08-17 09:12:46.306799+00:00", "lang": "en", "topics": ["large-language-models", "ai-products", "ai-research"], "entities": ["Alibaba", "Qwen 3.8 27B", "Simon Willison", "Claude Opus 4.6 Max", "Hacker News", "HuggingFace", "redashes/Qwen3.8-27B-BF16-SSMFIX"], "alternates": {"html": "https://wpnews.pro/news/qwen-3-8-27b-is-great-but-turn-off-its-brain-first", "markdown": "https://wpnews.pro/news/qwen-3-8-27b-is-great-but-turn-off-its-brain-first.md", "text": "https://wpnews.pro/news/qwen-3-8-27b-is-great-but-turn-off-its-brain-first.txt", "jsonld": "https://wpnews.pro/news/qwen-3-8-27b-is-great-but-turn-off-its-brain-first.jsonld"}}