YouTube2Text

Is Qwen 3.8 27b The New King of the Local LLMs? — Transcript

by Gary Explains · 1,101 words · 153 segments · language en · Watch on YouTube

Full transcript

  1. 0:00So, Qwen 3.8 27-billion parameter local
  2. 0:03AI model has been out for a few days
  3. 0:05now, and I've been testing it to see
  4. 0:07what it is capable of. And I've been
  5. 0:09surprised on how well it is performing,
  6. 0:11especially compared to the models that
  7. 0:13we had what, just a few weeks ago. This
  8. 0:16is another major step change in what you
  9. 0:19can achieve locally on local hardware
  10. 0:22without having to send everything up
  11. 0:24into the cloud. That's the topic of
  12. 0:25today's video. So, if you want to find
  13. 0:27out more, please let me explain.
  14. 0:32So, Qwen 3.8 is a 27-billion parameter
  15. 0:34model. Now, at its full size at 16-bit
  16. 0:38floating-point numbers, it's a 55-GB
  17. 0:41download, and you're probably going to
  18. 0:42need 128-GB of RAM to run it, which is
  19. 0:46beyond what most people have as VRAM,
  20. 0:49beyond what most people have kind of at
  21. 0:51home, even with a very good gaming
  22. 0:53setup. However, if you do have something
  23. 0:56like an RTX 5090 or something else with
  24. 0:5832-GB of VRAM, then the
  25. 1:02>> [snorts]
  26. 1:02>> 4-bit quantized version of Qwen 3.8
  27. 1:0527-billion is runnable on your local PC,
  28. 1:09and actually at a fairly decent rate. I
  29. 1:11was getting 96 tokens a second running
  30. 1:15it on an RTX 5090. That means you can
  31. 1:17run it locally on your PC and do
  32. 1:19everything you'd want to do with a
  33. 1:21normal large language model out in the
  34. 1:23cloud, but now you're running it
  35. 1:24locally. And I've been testing it
  36. 1:26specifically for how good is it at you
  37. 1:28doing coding using open code as the
  38. 1:32coding harness. But first, let's start
  39. 1:34with a few simple things. Does it answer
  40. 1:36the Alice question? Yes, it does, no
  41. 1:39problem, as you would expect. Does it
  42. 1:41handle the 7-minute egg timer and the
  43. 1:4411-minute egg timer question? Yes, it
  44. 1:47does, no problem. So, now those are out
  45. 1:50of the way, we can kind of get a feel
  46. 1:51for the kind of capabilities of this
  47. 1:53model, let's push it a a further. The
  48. 1:56first thing I did, cuz I wanted to have
  49. 1:57a bit of fun, was to get it to create a
  50. 1:59Space Invaders game using a single HTML
  51. 2:02file using inline JavaScript and no
  52. 2:05external libraries. So, this is just had
  53. 2:08to create everything itself inside the
  54. 2:10HTML file. It did that. It produced me
  55. 2:13index.html. You can double-click it and
  56. 2:15you get a nice Space Invaders game. So,
  57. 2:18I wanted to push it a bit more. So, I
  58. 2:19used my protocol encoding type length
  59. 2:22value encoding question. I asked it to
  60. 2:24create a specification similar to ASN.1,
  61. 2:28similar to Google's Protobuf. Now, it
  62. 2:30created that specification no trouble.
  63. 2:32There were a few tiny mistakes in it. It
  64. 2:34did all the specification defining how
  65. 2:36the protocol should be laid out. But,
  66. 2:38then at the very end it gives some
  67. 2:39working examples. This is what happens
  68. 2:41if you want to encode an integer. This
  69. 2:42is what happens if you want to encode a
  70. 2:43string. And actually one of those or two
  71. 2:46of those had a minor error in the
  72. 2:48numbers it was telling you that you
  73. 2:50should use. So, the examples it was
  74. 2:51giving uh didn't actually quite work
  75. 2:53out. But, if you fix those, no problem
  76. 2:55whatsoever. So, I then went on to say,
  77. 2:57"Well, now implement it including
  78. 2:58testing, including end-to-end testing,
  79. 3:00unit testing, and so on." And I did open
  80. 3:03code. Let it run. A few times I had to
  81. 3:05interact with it telling it to, you
  82. 3:07know, "Here's a plan. This is what you
  83. 3:09should do. Now, go ahead and do it." And
  84. 3:11just as you would do with normal coding
  85. 3:14with a harness. And it worked perfectly.
  86. 3:16It gave me a nice little library that
  87. 3:19could implement the specification
  88. 3:20including some end-to-end testing,
  89. 3:21including some unit tests. Worked
  90. 3:23absolutely fine. So, my big test after
  91. 3:26that, as I've done in several other
  92. 3:27videos here on this channel, is my new
  93. 3:30Scrippy uh programming language. So, the
  94. 3:32idea is I just give it a example of the
  95. 3:35of the new Scrippy language. I didn't
  96. 3:37define it formally. I tell it a little
  97. 3:39bit about it like it doesn't use
  98. 3:41semicolons. I tell it it's an
  99. 3:43interpreter. I tell it it's written in
  100. 3:45C. And then I say, "Go ahead. Knock
  101. 3:47yourself out. Go and write the
  102. 3:49interpreter for this language." Now,
  103. 3:51that's quite a tall order, because
  104. 3:52there's no formal specification for the
  105. 3:54language, just an example code, and a
  106. 3:57lot of the stuff it had to kind of infer
  107. 3:59from the example code, and infer from
  108. 4:01what I've said in the bit of text. Now,
  109. 4:04I've tried this before in other videos,
  110. 4:05and really local models weren't able to
  111. 4:08do it even Qwen 3.6. They weren't able
  112. 4:11to do it. They kind of either got stuck
  113. 4:13in a loop, just not being able to finish
  114. 4:15it, or what they produced crashed all
  115. 4:17the time, had to severe memory leaks. It
  116. 4:19was a bit of a disaster. However, the
  117. 4:21amazing news is here is that in a
  118. 4:24relatively short amount of time,
  119. 4:26certainly wasn't even overnight, just a
  120. 4:28few hours, again, few promptings, there
  121. 4:31was a plan, and then we went through the
  122. 4:33different steps on how to implement it,
  123. 4:35and then, you know, I said, "Also write
  124. 4:37some documentation." And I told it all
  125. 4:38the stuff back and forth as you would
  126. 4:40with a coding harness. It worked.
  127. 4:42Absolutely worked. You can run new
  128. 4:44Scrippy. You It created its own set of
  129. 4:47test scripts to run as well. Uh and the
  130. 4:51example script that I gave it for the
  131. 4:52readme file, that runs. And so, there
  132. 4:55you go. This local LLM running on a PC,
  133. 5:00Qwen 3.8, was able to do all of that
  134. 5:02coding using open code. You could use
  135. 5:04other harnesses as well, I'm sure.
  136. 5:06Haven't tested them specifically, but
  137. 5:07I'm sure it'll be just as good. And it
  138. 5:09did it all for me. And it actually
  139. 5:10works. So, this is really amazing that
  140. 5:12we've reached this point. Uh and I'm
  141. 5:14sure things are going to get better. But
  142. 5:17this is amazing that we've reached this
  143. 5:18point for local LLMs, just the power and
  144. 5:21the capabilities that they have. Okay,
  145. 5:23I'd love to hear your thoughts in the
  146. 5:24comments. Have you tried Qwen 3.8? Do
  147. 5:27you use a local AI for coding, or do you
  148. 5:31only rely on the big models? Do you not
  149. 5:33do any coding on it? Love to hear your
  150. 5:35thoughts in the comments below. Okay,
  151. 5:37that's it. I'll see you in the next one.
  152. 5:43>> [music]
  153. 5:58>> Ah.

About this transcript

This page contains the full transcript of Is Qwen 3.8 27b The New King of the Local LLMs? by Gary Explains, generated from the public captions YouTube serves with the video. The transcript has 1,101 words across 153 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to choose a caption track available on the video.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, caption language selection, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.