By April 2026, generative AI has more legitimate free courses than any other tech category. The problem isn’t access — it’s curation. After completing 22 free GenAI courses over the past year and tracking which ones actually moved my paid output (real LLM apps shipped, paid contracts won), here is the ranked list with realistic time commitments and what each one actually teaches.

Person studying AI on a laptop

Quick comparison

CourseProviderTimeLevelBest for
AI for EveryoneDeepLearning.AI6 hrsBeginnerNon-technical context
Generative AI for EveryoneDeepLearning.AI5 hrsBeginnerPractical mental model
Prompt Engineering for DevsDeepLearning.AI1 hrBeginnerFoundation prompting
LLM UniversityCohere12 hrsIntermediateRAG, fine-tuning basics
Hugging Face NLP CourseHugging Face25 hrsIntermediateOpen-source LLMs
Anthropic CookbookAnthropic15 hrsIntermediateClaude API, agents
Building with GeminiGoogle6 hrsBeginnerGoogle’s GenAI stack
Full-Stack LLM BootcampThe Full-Stack20 hrsAdvancedProduction LLM apps
Stanford CS25Stanford30 hrsAdvancedResearch-grade depth
AI Agents courseDeepLearning.AI10 hrsIntermediateLangGraph, CrewAI

1. AI for Everyone (Andrew Ng) — start here even if you’ve coded for 20 years

Andrew Ng’s classic non-technical introduction is still the best 6 hours you can spend understanding what AI can and can’t do. It won’t teach you to fine-tune anything, but it will fix mental models around training/inference, data quality, and project ROI.

  • Where: Coursera (free audit)
  • Strength: ROI thinking, project lifecycle, common pitfalls
  • Weakness: No coding, no LLM specifics

For full-time engineers, see how this pairs with AI Engineer Career Path 2026 for the broader skills map.

2. Generative AI for Everyone (Andrew Ng) — the GenAI version

Released in late 2024, refreshed for 2026. This is the GenAI-specific update of “AI for Everyone.”

  • Where: Coursera (free audit) / DeepLearning.AI
  • Strength: Realistic limitations, prompt design, business cases
  • Weakness: Light on code

A perfect companion to the technical courses below.

3. ChatGPT Prompt Engineering for Developers — the 1-hour high-leverage one

Andrew Ng + OpenAI. Free on DeepLearning.AI. Just one hour. Every developer should do it before writing a single LLM API call.

  • Where: learn.deeplearning.ai (free)
  • Strength: Concrete patterns (chain-of-thought, few-shot, output format)
  • Weakness: OpenAI-only examples (apply to Claude / Gemini with minor tweaks)

4. LLM University (Cohere) — RAG and fine-tuning fundamentals

Cohere’s free course covers retrieval-augmented generation, embeddings, and basic fine-tuning conceptually. It’s vendor-neutral enough to apply to OpenAI/Anthropic/Google.

  • Where: docs.cohere.com/llmu
  • Strength: Embedding theory, retrieval pipelines
  • Weakness: Cohere-flavored examples (still useful)

5. Hugging Face NLP Course — open-source path

If you want to own the model, not rent it, this is the course. 25 hours covers transformers, fine-tuning, and deployment with the Transformers library.

  • Where: huggingface.co/learn/nlp-course
  • Strength: Open-source tooling, parameter-efficient fine-tuning (PEFT/LoRA)
  • Weakness: Less focused on closed-source production patterns

Pair this with self-taught software engineer roadmap 2026 for foundational engineering skills.

6. Anthropic Cookbook — Claude API and agents

Free, code-first, official examples for tool use, prompt caching, computer use, and agentic workflows.

  • Where: github.com/anthropics/courses
  • Strength: Real production patterns (agents, tool use, Claude Code)
  • Weakness: Claude-specific (much transfers to other models)

The “Tool Use” and “Agents” sections are the two highest-leverage chapters in any free course.

7. Building with Gemini — Google’s GenAI stack

Covers Gemini API, Gemini in Vertex AI, multimodal inputs, and Google’s tooling.

  • Where: ai.google.dev / cloudskillsboost.google
  • Strength: Multimodal-first thinking, large context (2M tokens)
  • Weakness: Some Vertex AI material is paid Cloud-tier

8. Full-Stack LLM Bootcamp — production patterns

Originally a paid bootcamp, now released free. 20 hours covers LLMOps, evaluation, monitoring, and production deployment.

  • Where: fullstackdeeplearning.com (free)
  • Strength: Production realities — eval, monitoring, cost control
  • Weakness: Older recordings (2023–24) but principles still hold

9. Stanford CS25: Transformers United — research depth

Lecture series with talks from Geoffrey Hinton, Andrej Karpathy, OpenAI/Anthropic researchers. Free on YouTube.

  • Where: web.stanford.edu/class/cs25
  • Strength: Research-grade depth, latest architectures
  • Weakness: Time-intensive, requires ML foundations

10. AI Agents course (DeepLearning.AI) — agents in 2026

The 2026 production wave is agents. This series covers LangGraph, CrewAI, AutoGen, and ReAct patterns with hands-on labs.

  • Where: learn.deeplearning.ai
  • Strength: Practical agent patterns (planner-executor, multi-agent)
  • Weakness: Frameworks change quickly; concepts last longer

Suggested 90-day learning path

Week 1–2: Generative AI for Everyone + Prompt Engineering for Developers Week 3–4: LLM University (Cohere) modules 1–4 Week 5–6: Anthropic Cookbook + Building with Gemini Week 7–8: Hugging Face NLP Course modules 1–4 (open-source path) Week 9–10: Full-Stack LLM Bootcamp (production focus) Week 11–13: Build a real project applying everything

By day 90, you’ll have shipped at least one RAG app, one agent, and read modern transformer literature.

Cost reality — what’s actually free vs “free with caveats”

  • Truly free: DeepLearning.AI short courses, Hugging Face course, Anthropic Cookbook, Stanford CS25
  • Free audit, paid certificate: Coursera (Andrew Ng courses) — audits are free
  • Free with cloud charges: Google’s Vertex AI labs (small spend likely)
  • Free but server costs: Hugging Face GPU compute via free Spaces, or Colab free tier

Budget under $50 total to follow the entire path with optional GPU for fine-tuning experiments.

What about paid alternatives?

The two paid courses worth their money in 2026:

  • fast.ai Practical Deep Learning — pay-what-you-can; not strictly GenAI but the best engineering pedagogy
  • Coursera GenAI specializations with certificates — useful for resume signaling

For most learners, the free path above is more than enough. See Coursera vs Udemy 2026 for paid platform comparisons.

Common mistakes

  • Bingeing courses without building anything
  • Stopping at “AI for Everyone” — the application courses are where the value is
  • Following only one ecosystem (e.g., OpenAI-only) — diversify with Hugging Face open-source path
  • Skipping evaluation/monitoring (Full-Stack bootcamp covers this)
  • Falling for $999 “AI bootcamps” that repackage the same free content

Bottom line

You can become production-capable in generative AI for free in 2026. Start with Andrew Ng’s Generative AI for Everyone, follow with DeepLearning.AI’s prompt engineering and AI agents short courses, and end with the Anthropic Cookbook plus Full-Stack LLM Bootcamp for production-grade patterns. The market still pays a premium for engineers who have actually shipped LLM apps — and shipping is what these courses teach.

Sources

  • DeepLearning.AI course catalog April 2026
  • Hugging Face Learn statistics
  • Anthropic GitHub courses repository
  • Self-completed course logs and project outcomes (2025–2026)