Categories
Browse by what you’re building
Every model in the catalogue sits in exactly one category. Start from the capability you need and the shortlist follows.
- Language50General text generation and understanding
- Reasoning10Extended thinking for math, logic and analysis
- Coding4Code generation, review and agentic engineering
- Vision5Image and document understanding as input
- Image6Image generation as output
- Video5Video generation and understanding
- AudioSoonMusic and sound generation
- Speech2Text-to-speech and speech recognition
- Embeddings3Text-to-vector for search and RAG
- Agents1Tool use, computer use and multi-step autonomy
- Search1Retrieval-augmented and web-grounded models
- ResearchSoonDeep research and long-form synthesis
Why one category per model
Most models can be coaxed into most tasks, so a taxonomy that listed every model everywhere would be accurate and useless. Each model is filed under the capability it was built and priced for instead. A general language model appears under language even though it writes code; a model trained and benchmarked specifically for code appears under coding. The question the categories answer is not what a model can do, but what it is the right tool for.
Choosing where to start
Start from the output you need rather than the model you have heard of. If you need text back, begin with language or reasoning; images, with image; a transcript, with speech. Reasoning is worth its premium only where a task genuinely requires multi-step work — for extraction, classification and routing, a smaller language model is usually faster and an order of magnitude cheaper for the same result.