Qwen3.8 27B + DeepSeek Harness BEATS Opus 4.6 in Claude Code

Описание: Qwen3.8 27B + DeepSeek Harness takes on Claude Opus 4.6 in Claude Code. I run Qwen 3.8 27B locally on an RTX 5090 using Unsloth Desktop, compare its different reasoning levels through Tetris coding tests, and then connect the same local AI model to DeepSeek Harness for agentic coding. For the final challenge, Qwen3.8 27B and Opus 4.6 receive the same massive prompt to build a playable Minecraft clone. In my testing, this small 27B local LLM somehow produced the stronger result.

Deepseek Settings Command: notepad "$env:USERPROFILE\.dsh\settings.yaml"

models:
id: unsloth/Qwen3.8-27B-GGUF
name: Qwen3.8-27B
contextWindow: 200000
maxTokens: 131072
input: [text, image]
reasoningEfforts:
low: low
medium: medium
xhigh: xhigh

Minecraft Clone Prompt: https://docs.google.com/document/d/1g...

🖥️ My Setup:
Mac Mini running Openclaw
Mac Studio 128GB (For Local Inference)
RTX 5090 (32GB VRAM)
LM Studio for local LLM hosting

Unlock eternal happiness here ▶    / @james-layne  
X - https://x.com/jameslayneai

Cool Tech https://amzn.to/47yxnzB
Mac Mini https://amzn.to/4s3jpjB
Mac Studio https://amzn.to/4cidPEn
Nvidia 5090 https://amzn.to/483hXGJ

Purchases made through some links may provide some compensation to the creator.

Music Provided By: StreamBeats by Harris Heller Free Download: streambeats.com Listen on Spotify: search "StreamBeats"

00:00 - Qwen 3.8 27B
00:49 - Benchmarks
02:01 - Unsloth Desktop Setup
04:01 - Model Settings
05:54 - Qwen 3.8 Reasoning Level Testing
06:29 - No Reasoning Test
08:22 - Low Reasoning Test
09:54 - Medium Reasoning Test
11:32 - Extra High Reasoning Test
12:45 - Deepseek Harness
13:07 - Deepseek Harness Setup
15:03 - How To Use Deepseek Harness
17:57 - The Coding Challenge: Qwen 3.8 vs Claude Opus
19:11 - Qwen3.8 27B - Minecraft Clone
23:35 - Opus 4.6 - Minecraft Clone
28:08 - Verdict