The Big 6

The AI industry is dominated by six companies, each with a different philosophy:

Company	Flagship Model	Strength
OpenAI	GPT-5, Codex	Largest ecosystem, first-mover
Anthropic	Claude Opus 4.6	Safety-focused, longest context
Google	Gemini 2.5	Multimodal, search integration
Meta	Llama 4	Open source, community-driven
xAI	Grok 3	Real-time data, X integration
Alibaba	Qwen 3	Best cost/performance ratio

Each takes a fundamentally different approach to AI safety, openness, and pricing.

What is AGI?

Artificial General Intelligence — AI that can perform any intellectual task a human can.

Current AI is "narrow" — excellent at specific tasks but can't generalize:

• GPT-5 writes brilliant code but can't drive a car
• It can analyze a contract but can't cook dinner

The question isn't IF, but WHEN:

• Optimists: 2026-2028
• Realists: 2030-2035
• Skeptics: 2040+

Most experts agree: we're closer than anyone expected 5 years ago.

Model Benchmarks

How do we compare models objectively?

Key benchmark categories:

• Intelligence — reasoning, logic, knowledge (MMLU, ARC)
• Coding — code generation, debugging (HumanEval, SWE-bench)
• Math — mathematical reasoning (MATH, GSM8K)
• Speed — tokens per second, latency

Important: Benchmarks measure specific capabilities, not real-world usefulness. A model that scores lower on benchmarks might be better for YOUR use case.

Pricing Models

AI isn't free — understanding costs is critical:

Token-based pricing (most common):

• Pay per input token (your prompt) + output token (AI response)
• 1 token ≈ ¾ of a word
• Prices range from $0.04/1M (Nova Micro) to $75/1M (Opus output)

GPU hosting (self-hosted):

• Rent GPUs (A100, H100) — $2-8/hour
• Run open-source models (Llama, Qwen, Mistral)
• Free per token, but fixed infrastructure cost

The math: If you make >10,000 requests/day with the same model, self-hosting often saves 50-80%.

Hosting Options

Where should you run your AI?

Option	Pros	Cons
Cloud API (OpenAI, Anthropic)	Easy, no infrastructure, always latest models	Per-token cost, data leaves your network
Managed cloud (AWS Bedrock, Azure)	Enterprise compliance, your cloud account	Higher base cost, fewer models
Self-hosted (Ollama, vLLM)	Free per token, full privacy, air-gapped	GPU hardware needed, you maintain it
Hybrid	Best of both worlds	Most complex to manage

Pro tip: Start with cloud APIs, measure actual usage, then move high-volume workloads to self-hosted.

---quiz question: Which AI company is known for its open-source approach? options:

• { text: "OpenAI", correct: false }
• { text: "Anthropic", correct: false }
• { text: "Meta", correct: true }
• { text: "Google", correct: false } feedback: Meta releases its Llama models as open source, enabling the community to run, fine-tune, and deploy them freely.

---quiz question: What does AGI stand for? options:

• { text: "Advanced Generative Intelligence", correct: false }
• { text: "Artificial General Intelligence", correct: true }
• { text: "Automated GPT Interface", correct: false } feedback: AGI = Artificial General Intelligence — AI that can perform any intellectual task a human can. Current AI is still "narrow AI".

---quiz question: When is self-hosting AI models typically more cost-effective than cloud APIs? options:

• { text: "Always — self-hosting is always cheaper", correct: false }
• { text: "When you make more than ~10,000 requests/day with the same model", correct: true }
• { text: "Only for open-source models", correct: false } feedback: Self-hosting has a fixed GPU cost but zero per-token cost. The break-even depends on volume — typically around 10k+ requests/day.