Fastest Local Coding LLMs Tested: Nemotron 3 Nano vs Qwen 3.5 vs Devstral - Local AI Leveled Up

Author: Protorikis - Bewertung: 116x - Views:3467

Local inference capable LLMs are getting better and faster. Mac's unified memory opens up real local AI coding possibilities.

Here I test these models:
* Nemotron 3 Nano from NVidia
* Devstral Small 2 Instruct
* Qwen3.5 35B A3B
* Qwen3 Coder Next

I connect them to OpenCode AI coding agent and show what I found about their pros and cons, while implementing a utility internet speed test logging application.

Hardware for reference:
* MacBook Pro M3 Max 36GB

⏱️ Chapters

00:00 - Intro
00:17 - OpenCode
00:49 - Overview of the Models Tested
01:55 - The Goal Application to Produce (The Benchmark)
02:15 - Test 1/4: Nemotron 3 Nano
03:48 - Test 2/4: Devstral Small 2 Instruct
04:52 - Test 3/4: Qwen3.5 35B A3B (Amazing)
11:45 - Test 4/4: Qwen3 Coder Next (with extreme 1-bit quantization)
12:42 - Conclusion