Foundation model companies
Heavily funded. Enormous valuations.
The frontier models are fairly equivalent, with improvements arriving every few months. Pick one to pay for with all the features and learn how to drive it. Picking the right capabilities (deep research, search, etc.) and improving prompts, asking follow up questions, etc. will drastically improve results.
Individual model versions churn too fast to be worth memorizing. The durable structure: each company ships a family with a flagship tier, a balanced tier, and a fast/cheap tier. As of July 2026:
- OpenAI
- Models: the GPT-5 family, currently topped by the GPT-5.6 tier (Sol / Terra / Luna as flagship / balanced / fast)
- Products
- ChatGPT (web, mobile, desktop)
- Codex (coding agent)
- Anthropic
- Models: the Claude family. Claude Fable 5 is the most capable generally available model, with Opus (4.8), Sonnet (5) and Haiku (4.5) tiers below it
- Products
- Claude (web, desktop, mobile)
- Claude Code (CLI, desktop app, web, IDE extensions)
- Google
- Models: the Gemini family, currently the 3.x generation (Pro and Flash tiers)
- Products
- Gemini app (web, mobile)
- AI woven through Search and Workspace
- Meta
- Models: the Llama family (open-weight)
Consumer AI products
e.g. ChatGPT, Claude desktop, Gemini mobile
These consumer chat products are great for:
- Research and ideation
- Content generation - text and images
- Light code prototyping - make a fake web page design to show someone
- Very light actual app building, if you have a place to run the code (e.g. google apps script) or your own website.
- Typically a simple Javascript snippet or python script
Model companies make money from monthly subscriptions to these apps. Under the hood, their costs heavily depend on how much you use the product.
API access to base models
All of the base models can be accessed via APIs by developers, to be used in real time as part of their applications. Costs are per token and can be extremely high, depending on model and tuning.
Examples
- Converting images to text
- Summarizing user reviews
- Automated customer support that generates human sounding custom messages
- Fraud detection through analyzing unusual user behaviors
Model companies make money on a per token basis, so cost and revenue are more aligned.
AI product building platforms / vibe coding tools
These will build a website and host it for you, with limitations. They come with framework and design defaults that improve final results, but are too constraining for use in most large enterprises.
AI coding tools
Unlike vibe coding tools, these can help you build any code. You need an engineering background, or the patience to develop the instincts of one. These began as autocomplete assistants and are now fully agentic: the human writes a design spec, the AI builds it, and the human verifies the result.
- Anthropic’s Claude Code
- OpenAI Codex
- Cursor
- Github copilot