Loading video...

Video Failed to Load

Go Home

Build & Train a GLM-5.3-Flash Model From Scratch I built a 25.7M-parameter educational model inspired by GLM-5.3-Flash, then tested its architecture, pretraining, and executable-reward reinforcement learning end to end. Repository: YouTube: Become AI researcher in 90 days:

VukRosic99's profile picture

Vuk Rosić 武克

2,706 subscribers

14,590 views • 2 months ago •via X (Twitter)

Science & TechnologyEducation
LIVE

Anya Rossi• Live Now

Private livecam show

4 Comments

Maverick's profile picture
Maverick2 months ago

That's actually awesome

Yechan Do's profile picture
Yechan Do2 months ago

Pretty cool !

basedcapital's profile picture
basedcapital2 months ago

which glm-inspired component transferred least cleanly at 25.7m parameters, and what did you adapt for the educational model instead of preserving the architecture?

剑的 AI 实验室's profile picture
剑的 AI 实验室2 months ago

@VukRosic99 剑的 AI 实验室已核验 vukrosic/glm-5.3-flash-from-scratch，并收录到项目页： 。我们记录了 25.7M 参数教育模型和端到端训练说明，目前尚未实测；欢迎补充最小任务或纠正信息。

Related Videos

4 AI models trying to draw their self-portraits. GLM-5.3-Flash, GLM-5.3, Kimi K3, and Claude Opus 5. GLM-5.3-Flash came out looking the most handsome 🤓
0:50

4 AI models trying to draw their self-portraits. GLM-5.3-Flash, GLM-5.3, Kimi K3, and Claude Opus 5. GLM-5.3-Flash came out looking the most handsome 🤓

TypingMind

77,196 views • 2 months ago

Which flash model wins? DeepSeek v4.1 flash, Qwen 3.8 flash, or GLM 5.3 flash 🤯
0:41

Which flash model wins? DeepSeek v4.1 flash, Qwen 3.8 flash, or GLM 5.3 flash 🤯

GMI Cloud

15,052 views • 2 months ago

We use GLM-5.3-Flash build a dream kitchen. A 3D world built in Blender. This is not a generated video.
0:44

We use GLM-5.3-Flash build a dream kitchen. A 3D world built in Blender. This is not a generated video.

Lou

1,013,275 views • 2 months ago

Five frontier models. One simple prompt. No retries. No blender. Just a simple frontend web design. Fable 5.1 vs. GPT-6 Astra vs. DeepSeek v4.1 Flash vs. GLM 5.3 Flash vs. Qwen3.8 Flash Next Which model built the best looking luxury watch landing page? IMO Fable 5.1 and GLM 5.3 Flash look the best. LIVE DEMO:
0:43

Five frontier models. One simple prompt. No retries. No blender. Just a simple frontend web design. Fable 5.1 vs. GPT-6 Astra vs. DeepSeek v4.1 Flash vs. GLM 5.3 Flash vs. Qwen3.8 Flash Next Which model built the best looking luxury watch landing page? IMO Fable 5.1 and GLM 5.3 Flash look the best. LIVE DEMO:

Mia

16,546 views • 2 months ago

We gave GLM-5.3-Flash a Blender scene and 12 hours later it built this.
0:28

We gave GLM-5.3-Flash a Blender scene and 12 hours later it built this.

Lou

201,408 views • 2 months ago

GLM 5.3’s biggest leap came from post-training alone. “GLM 5.3 was built entirely by post-training the previous model. For me, that was a wake-up call.” “Post-training is far more GPU-efficient than pre-training from scratch. Plenty of companies can’t afford to train a model from scratch, but they have proprietary data. They can post-train themselves and beat some of the closed models.” “There’s still so much juice left to squeeze from post-training. We compared GLM 5.3’s cyber capabilities with 5.2, and it’s a night and day difference. That’s all just post-training.”
1:49

GLM 5.3’s biggest leap came from post-training alone. “GLM 5.3 was built entirely by post-training the previous model. For me, that was a wake-up call.” “Post-training is far more GPU-efficient than pre-training from scratch. Plenty of companies can’t afford to train a model from scratch, but they have proprietary data. They can post-train themselves and beat some of the closed models.” “There’s still so much juice left to squeeze from post-training. We compared GLM 5.3’s cyber capabilities with 5.2, and it’s a night and day difference. That’s all just post-training.”

SemiAnalysis

214,839 views • 27 days ago

GLM 5.3 Flash on 2x RTX PRO 6000s
0:23

GLM 5.3 Flash on 2x RTX PRO 6000s

Mia

58,307 views • 10 days ago

GLM 5.3 Flash vs Hybrid (GLM 5.3 Flash + Jev) on a turn-based strategy game (Halite 2) The hybrid approach is 13x faster, 56% of the pure GLM 5.3 Flash API cost, and gains a slight edge in performance. In the hybrid approach: -GLM 5.3 Flash handles strategic judgment of which planets matter, who expands, who fights, when to re-plan. -Jev handles the ship-by-ship execution. GLM 5.3 Flash alone defeats Jev alone 82% of the time. This is obviously just a single example and some of the first things I've tried to make use of Jev as an augmentation to an LLM. Could potentially make even more gains in cost, speed, and performance, but using what I've figured out here, I am compelled to try some ideas with robotics next.
0:24

GLM 5.3 Flash vs Hybrid (GLM 5.3 Flash + Jev) on a turn-based strategy game (Halite 2) The hybrid approach is 13x faster, 56% of the pure GLM 5.3 Flash API cost, and gains a slight edge in performance. In the hybrid approach: -GLM 5.3 Flash handles strategic judgment of which planets matter, who expands, who fights, when to re-plan. -Jev handles the ship-by-ship execution. GLM 5.3 Flash alone defeats Jev alone 82% of the time. This is obviously just a single example and some of the first things I've tried to make use of Jev as an augmentation to an LLM. Could potentially make even more gains in cost, speed, and performance, but using what I've figured out here, I am compelled to try some ideas with robotics next.

Harrison Kinsley

54,330 views • 1 month ago

DeepSeek v4.1 Flash vs GLM 5.3 Flash I asked both models to recreate the Psychonauts 2 and Metroid Prime game menus as closely as possible to the originals. Attached are full-sized videos from both models for both games. All were done in a single shot, both at Max thinking, same harness. Tokens used: DeepSeek v4.1 Flash - Psychonauts 2: 412k - Metroid Prime: 466k GLM 5.3 Flash - Psychonauts 2: 134k - Metroid Prime: 111k That's between 3-4x MORE used by DS4.1F. Who do you think won? IMO GLM 5.3 Flash won by FAR on both.
0:20

DeepSeek v4.1 Flash vs GLM 5.3 Flash I asked both models to recreate the Psychonauts 2 and Metroid Prime game menus as closely as possible to the originals. Attached are full-sized videos from both models for both games. All were done in a single shot, both at Max thinking, same harness. Tokens used: DeepSeek v4.1 Flash - Psychonauts 2: 412k - Metroid Prime: 466k GLM 5.3 Flash - Psychonauts 2: 134k - Metroid Prime: 111k That's between 3-4x MORE used by DS4.1F. Who do you think won? IMO GLM 5.3 Flash won by FAR on both.

Mia

24,127 views • 1 month ago

Jumpgate - hyperspace jump DS4.1F-1017 vs GLM 5.3 Flash vs Fable 5.1 Who won imo: 🥇 Fable 5.1 🥈 DeepSeek V4.1 Flash 1017 🥉 GLM 5.3 Flash You can run these by yourself here: Prompt below.
0:24

Jumpgate - hyperspace jump DS4.1F-1017 vs GLM 5.3 Flash vs Fable 5.1 Who won imo: 🥇 Fable 5.1 🥈 DeepSeek V4.1 Flash 1017 🥉 GLM 5.3 Flash You can run these by yourself here: Prompt below.

Mia

13,071 views • 25 days ago

GLM 5.3 free on Kira AI, 100M tokens + 50M check-in 😳 GLM 5.3 + Flash + Qwen 3.8 Flash free. 1M context, no card. → → sign up → grab key → plug in Third-party proxy, don't send secrets. Go grab it.
0:27

GLM 5.3 free on Kira AI, 100M tokens + 50M check-in 😳 GLM 5.3 + Flash + Qwen 3.8 Flash free. 1M context, no card. → → sign up → grab key → plug in Third-party proxy, don't send secrets. Go grab it.

Punch AI

17,124 views • 2 months ago

GLM 5.3 Flash EXL3 running on 3x DGX Sparks.
0:27

GLM 5.3 Flash EXL3 running on 3x DGX Sparks.

Mia

12,535 views • 2 months ago

Qwen 3.8 vs GLM 5.3 vs Gemini 3.8 Flash vs DeepSeek v4.1 Flash Prompt: Build an Angry Birds clone Qwen 3.8 took $0.49 in 3 min GLM 5.3 took $0.46 in 2 min Gemini 3.8 Flash took $0.29 in 2 min 23 sec DeepSeek v4.1 Flash took $0.11 in 1 min 28 sec Which one wins for you?
0:24

Qwen 3.8 vs GLM 5.3 vs Gemini 3.8 Flash vs DeepSeek v4.1 Flash Prompt: Build an Angry Birds clone Qwen 3.8 took $0.49 in 3 min GLM 5.3 took $0.46 in 2 min Gemini 3.8 Flash took $0.29 in 2 min 23 sec DeepSeek v4.1 Flash took $0.11 in 1 min 28 sec Which one wins for you?

Bhavy☄️

27,799 views • 2 months ago

Meet Wally. Our inference stack for open frontier models, built to be the fastest place to run them. Performance snapshot: GLM-5.3 Flash: 380 tok/s. GLM-5.3 Max: 790 tok/s Qwen3.8-27B: 485 tok/s DeepSeek-V4.1 Flash: 615 tok/s 1/7
0:46

Meet Wally. Our inference stack for open frontier models, built to be the fastest place to run them. Performance snapshot: GLM-5.3 Flash: 380 tok/s. GLM-5.3 Max: 790 tok/s Qwen3.8-27B: 485 tok/s DeepSeek-V4.1 Flash: 615 tok/s 1/7

RunAnywhere

66,857 views • 1 month ago

We are all sitting in a room watching GLM-5.3-Flash write Blender script for automotive museum.
1:00

We are all sitting in a room watching GLM-5.3-Flash write Blender script for automotive museum.

Lou

44,446 views • 2 months ago

fixed the reasoning bugs + the thinking loop in glm-5.3 flash. it's a different model now. i gave the same prompt to glm (running on my desk) vs gpt-6.1 sol. they produced a SINGLE HTML file each. very content with GLM's version, i'm genuinely impressed
0:41

fixed the reasoning bugs + the thinking loop in glm-5.3 flash. it's a different model now. i gave the same prompt to glm (running on my desk) vs gpt-6.1 sol. they produced a SINGLE HTML file each. very content with GLM's version, i'm genuinely impressed

filipe

61,707 views • 1 month ago

GLM 5.3 Flash vs DeepSeek V4.1 Flash 1017 Can't believe we're abusing gummy bears to test models, but here we are! Both models were run locally. - GLM 5.3 Flash: my own EXL3 quant (link below) - DeepSeek V4.1 Flash 1017: official quant - Same harness.Both on Max effort. Who won?
0:33

GLM 5.3 Flash vs DeepSeek V4.1 Flash 1017 Can't believe we're abusing gummy bears to test models, but here we are! Both models were run locally. - GLM 5.3 Flash: my own EXL3 quant (link below) - DeepSeek V4.1 Flash 1017: official quant - Same harness.Both on Max effort. Who won?

Mia

32,512 views • 27 days ago

NVIDIA is giving away 9 frontier AI models for free 🤯 • DeepSeek V4.1 Flash • Kimi K3 + K2.6 • GLM 5.3 + GLM 5.3 Flash • Nemotron 3 Ultra 550B • Nemotron 3 Super 120B • Nemotron 3.5 Lightning • Nemotron 3 Nano Omni No card, 40 req/min. Base URL:
0:20

NVIDIA is giving away 9 frontier AI models for free 🤯 • DeepSeek V4.1 Flash • Kimi K3 + K2.6 • GLM 5.3 + GLM 5.3 Flash • Nemotron 3 Ultra 550B • Nemotron 3 Super 120B • Nemotron 3.5 Lightning • Nemotron 3 Nano Omni No card, 40 req/min. Base URL:

Punch AI

17,655 views • 1 month ago

We're proud to be the fastest inference provider on Artificial Analysis, OpenRouter, and Hugging Face for GLM-5.3-Flash, at 122+ TPS. All served from the US only, starting on day 0, with ZDR by default. Stay tuned for updates as our engineers continue to optimize GLM-5.3-Flash for even higher throughput and lower latency.
0:20

We're proud to be the fastest inference provider on Artificial Analysis, OpenRouter, and Hugging Face for GLM-5.3-Flash, at 122+ TPS. All served from the US only, starting on day 0, with ZDR by default. Stay tuned for updates as our engineers continue to optimize GLM-5.3-Flash for even higher throughput and lower latency.

Baseten

23,114 views • 2 months ago

How prose looks like in Hermes agent using GLM 5.3 Flash EXL3 on 2x DGX Sparks
0:30

How prose looks like in Hermes agent using GLM 5.3 Flash EXL3 on 2x DGX Sparks

Mia

10,574 views • 2 months ago