Run Open-Weight Models for Free: DeepSeek, Qwen, and GLM on ModelScope Tutorial
Sign up with my invite code jonlaw at: https://modelscope.ai/register?inviteCode=jonlaw or comment "modelscope".
Open-source AI models like Qwen, DeepSeek, GLM, and Kimi now compete with GPT, Claude, and Gemini. But testing them usually means three API accounts, three keys, and a bill. In this video, I use ModelScope's free API-Inference to run DeepSeek V4 Pro, Qwen 3.8, and GLM 5.2 head-to-head. I test each model on my own coding, reasoning, and writing prompts through one API, with no credit card. Then I take the winner and build a customer support email app for Alto, the coffee brand from my past videos.
I also show you how to filter the ModelScope catalog for free API-inference models, get your access token, and run a multi-model comparison script. At the end, I give my honest take on the limitations. That includes the Magic Cubes credit system, translation gaps, and free tier quotas.
CHAPTERS:
0:00 Open-Source AI Models Go Mainstream (Qwen, DeepSeek, GLM, Kimi)
0:31 The Problem With Testing Open-Source Models Across API Providers
0:51 Head-to-Head Model Test on One Free API
1:22 What ModelScope Is Used For in This Video
1:53 Why Open-Source AI Matters in 2026
2:38 Benchmark Cherry-Picking and Training Bias
3:00 Why Benchmarks Don't Answer Your Use Case
3:14 API Sign-Ups vs Running Models Locally
3:41 What is ModelScope? The Hugging Face Alternative
4:22 Free API Inference, GPU Notebooks, Storage, and App Deployment
4:44 ModelScope Magic Cubes Explained
5:03 Get 100 Extra Magic Cubes With Invite Code
5:24 The Test Plan: One Use Case, One Free API
5:36 New ModelScope Account With No Credit Card
5:53 Browsing the ModelScope Model Catalog
6:34 How to Filter for Free API-Inference Models
7:00 Inside a ModelScope Model Page
7:32 Getting the API Inference Code and Web Demo
7:54 How to Get Your ModelScope Access Token
8:13 Setting Up the Open-Source Model Showdown
8:44 DeepSeek V4 Pro vs Qwen 3.8 vs GLM 5.2
8:53 The Python Script for Multi-Model Comparison
9:42 Running Coding, Reasoning, and Writing Tests in Terminal
10:17 Results: Speed and Token Usage Compared
10:48 Coding and Reasoning Winner: DeepSeek
11:05 Writing Winner: Qwen 3.8
11:19 Building an App With the Winning Model
11:37 Demo: AI Customer Support Email App With DeepSeek
12:37 Recap: 3 Frontier Models Tested for $0
12:57 Limitation 1: The Magic Cubes Credit System
13:25 Limitation 2: Translation Gaps on the International Site
13:42 Limitation 3: Not Every Model Has API Inference
13:59 Limitation 4: Free Tier Quotas for Production Apps
14:26 Who Should Use ModelScope
15:03 How to Get Started With ModelScope for Free
TAGS:
modelscope, modelscope api, modelscope tutorial, modelscope free api, modelscope api inference, hugging face alternative, free llm api, free ai api no credit card, open source ai models, open source llm comparison, deepseek v4 pro, qwen 3.8, glm 5.2, deepseek vs qwen, deepseek vs qwen vs glm, best open source llm 2026, free deepseek api, free qwen api, compare ai models, test llm models, open weight models, magic cubes modelscope, build ai app free, python llm api