Llama, Mistral and Qwen Overview
The three big open-weight model families: licenses, sizes, and what each is good at.
Llama, Mistral and Qwen Overview is a free AI Agents lesson on CoddyKit — lesson 1 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the AI Agents learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.
Why Open Source?
Open-weight models give you:
- No per-token API cost
- Data privacy — model runs on your hardware
- Customization (fine-tuning) at any scale
- No vendor lock-in
Cost: engineer hours, GPU bills, and usually worse quality at the frontier.
Llama Family (Meta)
- Released 2023-2025 in sizes 1B to 405B
- Llama 3.x is highly capable; 8B competes with mid-tier closed models
- Community license (permits commercial use under conditions)
- Huge ecosystem; most fine-tunes are Llama-derived
Mistral Family (Mistral AI)
- Mistral 7B Instruct — strong small model
- Mixtral 8x7B — Mixture-of-Experts, fast and capable
- Mistral Large / Medium — closed weight, available via API
- Apache 2.0 license on open variants
Qwen Family (Alibaba)
- Qwen 2.5 — top open-weight reasoning model (2024)
- Excellent multilingual support, especially Chinese
- Sizes from 0.5B to 72B
- Apache 2.0
Other Notables
- Gemma (Google) — small, polished
- Phi (Microsoft) — tiny but capable
- DeepSeek — strong reasoning at lower cost
- Yi — strong multilingual
Picking a Size
| Size | Use case |
|---|---|
| 1-3B | Edge devices, mobile, classification |
| 7-8B | Single GPU, simple agents |
| 13-30B | Bigger GPU or 2-4 small GPUs, real agents |
| 70B+ | Multi-GPU, frontier quality |
Benchmarks That Matter
- MMLU — general knowledge
- GSM8K — math
- HumanEval — code
- MT-Bench — chat quality
- HELM / LMSYS Chatbot Arena — human preference
Pick benchmarks aligned with YOUR task.
Frontier Gap Is Closing
Frontier closed models (GPT-4o, Claude Sonnet 4.5) are still ahead on the hardest benchmarks. But open 70B models match closed mid-tiers on most agent tasks.
License Considerations
Check every model's license:
- Apache 2.0 — fully permissive
- Llama Community License — restrictions on huge-scale services and certain use cases
- Some "research only" licenses — cannot deploy commercially
Hosted vs Self-Hosted
You can use open models via hosted APIs (Together AI, Anyscale, Groq, Fireworks) without running infrastructure — often cheaper than OpenAI for the same quality.
Hosted Open Models on Groq
from openai import OpenAI
client = OpenAI(
base_url='https://api.groq.com/openai/v1',
api_key=GROQ_KEY
)
response = client.chat.completions.create(
model='llama-3.1-70b-versatile',
messages=[{'role': 'user', 'content': 'Hello'}]
)When OSS Wins
- Privacy-critical: medical, legal, defense
- Very high volume: cents-per-call matters
- Heavy customization: fine-tunes for a specific domain
- Multilingual where closed providers are weak
When Closed Models Win
- Frontier reasoning
- Multimodal (vision, voice)
- Long-context: Claude / Gemini still lead
- Quick prototyping
Smallest Capable Model
For a simple internal agent that runs on a single GPU, which size range is typical?
Recap
Llama, Mistral, Qwen are the big three. Pick the smallest size that meets quality. Hosted providers (Groq, Together) let you avoid infrastructure.
Frequently asked questions
Is the “Llama, Mistral and Qwen Overview” lesson free?
Yes — the full text of “Llama, Mistral and Qwen Overview” is free to read here on the web, and the AI Agents course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the AI Agents course, upgrade to CoddyKit PRO.
What will I learn in “Llama, Mistral and Qwen Overview”?
The three big open-weight model families: licenses, sizes, and what each is good at. You practise AI Agents with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.
Do I need any experience to start AI Agents?
No prior experience is required. AI Agents on CoddyKit is structured for beginners through advanced learners; this is — lesson 1 of 4, so you can start here or from the beginning and move at your own pace.
How long does the “Llama, Mistral and Qwen Overview” lesson take?
Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.
Can I write and run code in this AI Agents lesson?
Yes. Every AI Agents lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.