I Was WRONG About Local LLM vs Claude on M5 macbook pro — TESTED | qwen 3.6 35b, 27b, and gtp oss

Can a $2400 MacBook Pro M5 actually replace Claude? I put three local
LLMs — Qwen3.6-35B-A3B, Qwen3.6-27B, and gpt-oss-20b — head-to-head
against Claude Sonnet 4.6 inside VS Code. The results were not what I
expected.

In this video I run all three open-weight models locally in LM Studio
on the base M5 MacBook Pro, benchmark real tokens-per-second, then ask
each one to build a solar system website. Finally, I open the broken
project in VS Code and let Claude Sonnet 4.6 (via GitHub Copilot) fix
it — and the difference is wild.

If you're a developer thinking about running local AI on Apple Silicon,
a Mac buyer wondering whether the M5 is enough for LLM run locally, or you just
want to see where open models stand against Anthropic's best, this one
is for you.

⏱️ CHAPTERS
🔗 TOOLS USED
- LM Studio (run LLMs locally on Mac): https://lmstudio.ai/
- Ollama (alternative local runner): https://ollama.com/
- Visual Studio Code: https://code.visualstudio.com/
- GitHub Copilot: https://github.com/features/copilot

🤖 MODELS TESTED
- Qwen3.6-35B-A3B (Apache 2.0, Alibaba):
https://huggingface.co/Qwen/Qwen3.6-35B-A3B
- Qwen3.6-27B (Apache 2.0, Alibaba):
https://huggingface.co/Qwen/Qwen3.6-27B
- gpt-oss-20b (Apache 2.0, OpenAI):
https://huggingface.co/openai/gpt-oss-20b
- Claude Sonnet 4.6 (Anthropic):
https://www.anthropic.com/claude/sonnet

💻 MY MACHINE
MacBook Pro M5 chip
14 inch
32gb unified memory (ram)
10 core cpu
10 core gpu
1 tb ssd
(Released Oct 22, 2025 — starting at $1,599)
Apple newsroom: https://www.apple.com/newsroom/2025/10/

📝 NOTES
Qwen3.6-35B-A3B is a sparse Mixture-of-Experts model — only ~3B of
its 35B parameters activate per token, which is exactly why it runs
6–7x faster than the dense Qwen3.6-27B on the same Mac. If you've
been wondering why "bigger" models sometimes feel snappier, that's
the secret.

👋 ABOUT ME
I'm Vikramjit. This channel is where I test AI tools, run local LLMs
on Apple Silicon, and document what actually works for developers.

🔔 If you got value from this, please Like, Subscribe, and tell me in
the comments which model YOU want me to test next. I read every comment.

📬 SOCIALS / CONTACT
- YouTube: https://www.youtube.com/@IamVikramjit
#LocalLLM #MacBookProM5 #Qwen3 #gptoss #ClaudeSonnet #LMStudio
#AppleSilicon #AIcoding #Ollama #OpenSourceAI

Видео I Was WRONG About Local LLM vs Claude on M5 macbook pro — TESTED | qwen 3.6 35b, 27b, and gtp oss канала IamVikramjit