Skip to main content

The AI Model Landscape: Choosing Among Claude, ChatGPT, Gemini & Open Models

Beginner

AILmanac's core is Claude — but the skills travel. The same prompting, context-engineering, tool-use and eval habits work on almost any modern model. This page widens the lens to the whole field: who the major players are, how to choose for a task, and what carries over no matter which you use.

What you'll learn
  • Know the major AI assistants & models and what archetype each one is
  • Use a repeatable framework to pick a model for a specific task — not by hype
  • See which skills transfer across every model (most of them)
  • Know which facts go stale fast — and where to check the current truth

The major players (by archetype, not ranking)​

Think in archetypes — the rankings shuffle, the archetypes are stable:

  • Anthropic — Claude · frontier closed model. Known for coding, agentic/tool use, long-context reasoning, and a safety-first posture. (The whole rest of this site goes deep here.)
  • OpenAI — ChatGPT / GPT · frontier closed model with the largest consumer ecosystem and a very broad, multimodal feature set.
  • Google — Gemini · frontier closed model, deeply wired into Google's products and data, with very large context windows.
  • Meta — Llama · open-weight: you can download and self-host it. The backbone of much of the local/open ecosystem.
  • Mistral · open-weight + API, European, efficiency-focused.
  • xAI — Grok · closed, integrated with X, leaning on real-time signal.
  • DeepSeek / Qwen (Alibaba) · strong open-weight models known for capability-per-cost; popular for self-hosting.
Pro tip
  • The biggest practical fork is CLOSED frontier (best raw capability, hosted only) vs OPEN-WEIGHT (self-hostable, private, customizable, often cheaper to run at scale).
  • Don't shop by leaderboard. The only benchmark that matters is YOUR task on YOUR data.

How to choose a model for a task​

Guided walkthrough1 of 6
  1. What must it do, and what's non-negotiable: data privacy (can data leave your network?), budget, latency, max context, regulated domain, on-device requirement.

Provider-neutral system prompt skeleton (works on any model)

You are a {role}. 
Goal: {one sentence}.
Rules:
- Use ONLY the provided context; if the answer isn't there, say "I don't know".
- Output: {exact format / schema}.
Context:
{context}
Task: {task}

What transfers across (almost) every model​

The reason learning Claude well pays off everywhere — these skills are portable:

  • Prompting structure — clear role, explicit task, examples, output format → Prompting Basics
  • Context engineering — deciding what goes in the window → Foundations
  • Tool use / function calling — the pattern is near-identical across providers → Tool Use
  • Retrieval (RAG) — ground answers in your data → RAG
  • Evals — measure quality so you can compare models honestly → Evals
  • Structured output — make results machine-readable → Structured Output

What does NOT transfer cleanly: exact API shapes, tool-calling JSON details, token limits, pricing, and the safety/refusal behavior — always re-check per provider.

Landscape vocabulary
Press Enter or Space to flip the card. Use the left and right arrow keys to move between cards.Term shown.
1 / 4

Check yourself

0/3
  1. What is the single most reliable way to choose between two models for YOUR task?
  2. You need to process sensitive data that cannot leave your network. Which way do you lean?
  3. Which of these does NOT transfer cleanly between providers?
Key takeaways
  • Think in archetypes (closed-frontier vs open-weight); the rankings churn, the archetypes don't.
  • Choose by constraints + a tiny eval on your own data — never by leaderboard hype.
  • Most skills (prompting, context, tools, RAG, evals) transfer across every model — that's why going deep on Claude pays off everywhere.
  • Names, prices, limits and benchmarks go stale fast — verify current specifics at the source.

Sources & further reading​

Next​