TL;DR: Andrej Karpathy’s Neural Networks: Zero to Hero is a free ~15-hour video series in which one of the world’s best AI educators builds neural networks — and eventually a working GPT — entirely from scratch, in code, on camera. It’s the most revered deep-learning resource on the internet. But fair warning: despite the name, it’s not for total beginners. You need solid Python and comfort with calculus basics before pressing play.
What is Neural Networks: Zero to Hero?
It’s a lecture series by Andrej Karpathy — former senior director of AI at Tesla, founding member of OpenAI, Stanford PhD, and founder of the AI-education company Eureka Labs. The series starts at the absolute root: hand-coding backpropagation through a tiny autograd engine he calls micrograd. From there it builds a character-level language model (makemore), step by step, into MLPs with batch normalization, then a WaveNet-style network — and ends with the famous two-part finale: building GPT from scratch (based on the “Attention Is All You Need” paper) plus a separate video on building a GPT tokenizer with byte-pair encoding.
All coding is done live in Python and PyTorch, with companion code and notes on GitHub at karpathy.ai/zero-to-hero.html. The full series runs about 15 hours across roughly 10 videos.
Pricing (verified October 2026)
| Item | Price | Notes |
|---|---|---|
| All videos on YouTube | $0 | Free, no account needed |
| Companion code and notes | $0 | Open on GitHub |
| Certificate | Not offered | No credential — pure skill-building |
| Compute | $0 | Early videos run on a laptop or free Colab; the GPT-2 reproduction video is watch-only for most people (needs serious GPUs) |
What’s good
- The best teacher in the business. Karpathy explains backprop, batch norm, and attention with a clarity no textbook matches. Concepts that take chapters elsewhere click in minutes here.
- You build everything from nothing. No frameworks doing the magic for you at first — you implement autograd, MLPs, and transformers by hand. That first-principles understanding never goes stale.
- The GPT finale is unmatched. Watching a working GPT get built from scratch, line by line, demystifies LLMs in a way no prompting course can.
- Completely free, no strings. Arguably the highest-quality deep-learning education on earth, and it costs nothing.
- Active community. Thousands of learners have worked through it, so study notes, companion repos, and forum answers exist for every hard part.
What’s not so good
- “Zero to Hero” is aspirational. Class Central rates it advanced, and they’re right. Solid Python and comfort with derivatives and basic probability are real prerequisites. Absolute beginners will drown.
- No certificate or credential. Employers won’t see a badge; the value is entirely in what you can now do (and demonstrate in interviews or projects).
- Time-intensive. Each ~2-hour lecture easily becomes 5–6 hours when you code along, debug, and rewatch the dense parts. Plan accordingly.
- No exercises or quizzes. It’s a lecture series, not a course platform. Discipline and self-testing are on you.
Who it’s best for
Programmers — ideally with some NumPy/PyTorch exposure — who want deep, first-principles understanding of neural networks and transformers. Perfect as a second course after a gentler intro (fast.ai’s Practical Deep Learning, Microsoft’s AI for Beginners, or StatQuest videos). Not for non-programmers, and not for anyone who just wants to use AI tools at work.
The verdict
Neural Networks: Zero to Hero is the single best free deep-learning resource ever made — if you meet the prerequisites. Nothing else takes you from “what is backprop” to “I built GPT” in 15 hours of video. Just don’t let the friendly title fool you: get comfortable with Python and basic calculus first, code along with every video, and you’ll come out understanding neural networks at a level most paid courses never reach.
FAQ
Is Neural Networks: Zero to Hero really free?
Yes — every video is free on YouTube, and all companion code is open on GitHub. There’s no paid tier.
Is Zero to Hero good for beginners?
It’s excellent for programming beginners-to-intermediate who want deep learning specifically — but not for total beginners. You need solid Python and basic calculus/derivatives first.
How long does Zero to Hero take?
About 15 hours of video, but most learners spend 30–50+ hours total when coding along and working through the dense material.
Do I get a certificate?
No. There’s no certificate — the payoff is the skill itself, which you can demonstrate through projects or interviews.
Prices checked 2026-10-09 and can change — confirm on the platform’s site before buying.