Learning path

The open-weights progress arc

Open-weight AI has gone from a niche curiosity to a genuine rival of closed frontier models — in just a few years. This path traces that arc: the infrastructure that made sharing models easy, the labs and companies driving releases, and the flagship open models that shifted the field's expectations. For readers who want to understand who's building what, and why open weights matter.

Mixed level8 steps~56 min

8 steps

Begin →

Hugging Face
Start here: Hugging Face is the platform that made open-weight models shareable and usable at scale — the infrastructure the rest of this path runs on.
Read →Beginner In-depth
Meta
Meta's Llama releases were the first major signal that a top-tier lab would open its weights — the move that legitimized the whole ecosystem.
Read →Beginner In-depth
Mistral AI
Mistral AI showed that a small European lab could punch well above its weight class with lean, open models — broadening who could compete.
Read →Beginner In-depth
Alibaba
Alibaba's backing of Qwen established that Chinese tech giants were serious open-weight players, not just observers.
Read →Beginner In-depth
Qwen
Qwen itself — Alibaba's prolific model family — is one of the most active open-weight release lines, spanning sizes from small to frontier.
Read →Beginner In-depth
DeepSeek V4
DeepSeek V4 is the current benchmark for what open weights can achieve at the frontier, and the release that most recently reset expectations for the field.
Read →Beginner In-depth
NVIDIA
NVIDIA's hardware and software stack is the engine underneath nearly every open-weight training and inference run — understanding it grounds the economics of open AI.
Read →Beginner In-depth
Interconnects
Interconnects covers the networking infrastructure that makes large-scale open-weight training feasible — the often-overlooked physical layer of the stack.
Read →Beginner In-depth

Hugging Face

Meta

Mistral AI

Alibaba

Qwen

DeepSeek V4

NVIDIA

Interconnects