AI model landscape 2026
About this app
Compare major AI model families by capability, reasoning depth and openness in a 3D bubble chart. Bubble size represents context length; a qualitative snapshot without version numbers.
The idea
Large language models differ along several dimensions at once: how fast and cheaply they respond, how well they reason over several steps, how much text they can handle at once (the context window) and whether their weights are openly available. Open weights allow local deployment and custom fine-tuning, while proprietary models usually run through a provider’s API. Because individual versions change quickly, grouping by model family is more stable than ranking single models.
What you see
A 3D bubble chart with seven families: GPT, Claude, Gemini, Llama, Mistral, DeepSeek and small local models. The x axis runs from fast and cheap to powerful and expensive, the y axis from broad tasks to deep reasoning, and the z axis from open to proprietary weights. Bubble size stands for the context window and colour for the provider. Each family has a description, strengths and typical uses. The positions are a qualitative snapshot from mid-2026 without version numbers.
Things to try
Rotate the scene and click a bubble, or step through them with “Next model”. Compare which families have open weights and where small local models sit. Read the axes as rough orientation: in practice strengths and weaknesses overlap considerably, and no benchmark ranking can be derived from the chart. Each family’s card lists strengths and typical uses.
Frequently asked questions
- What is a context window in AI models?
- The context window is the maximum amount of text, measured in tokens, that a language model can take into account in one request, including the input and the conversation so far. A larger window allows long documents or a lot of code to be processed at once.
- What does “open weights” mean for a language model?
- For open-weight models the trained parameters can be downloaded. You can then run the model on your own hardware and fine-tune it. This does not automatically mean that training data or training code are public, and licence terms vary from model to model.
- What are reasoning models?
- Reasoning models are language models that generate several intermediate steps before answering, solving complex tasks step by step, for example in maths or programming. This usually takes more compute per request.
- What can I do in this app?
- You rotate a 3D bubble chart of seven model families, select a bubble by clicking or with “Next model”, and read provider, openness, context window, strengths and typical uses. The placement is qualitative and deliberately has no version numbers.
Subject: Machine Learning
More from Machine Learning
- AdaBoost
- Anatomy of a prompt
- Data privacy traffic lights
- Decision Tree
- Ensemble: Bagging vs. Boosting
- Finding faces: masks versus learning from examples
- k-Means
- k-Nearest Neighbors (kNN)
A VisuApp by heyprof: interactive, free in your browser, no sign-up. What is a VisuApp? · All apps