Skip to main content
Shipped

In the wild, and mapped as it moves.

The map

How the field is taking shape

Signals
70
Entities
282
Clusters
4

Table: every model we track. Graph: how concepts connect. River: how clusters change week to week.

Table

The periodic table of what is in the wild

Every model we track, as an element. Columns are what it does; rows are how big it is. Where a cell is empty, nothing we track has shipped there yet.

The table fills in as signals are tagged with their model details.

Decide
Models whose job is a judgement, not prose
Route
Models that send a request to the right model or tool
Guard & judge
Safety classifiers and judges
Speak
Speech and text-to-speech
Language
General language models
Code
Coding models
See
Vision-language models

How to read a tile

  • Thick solid border · open licence stated (Apache-2.0, MIT)
  • Dashed border · non-commercial or research-only licence
  • Thin border · licence not stated, or custom terms
  • Available
  • Verified
  • Announced
  • Symbol · two letters, unique across the table, and stable once assigned.
  • Size · Rows: total parameters. Mixture-of-experts shows total / active.
  • On X · built on another model in the table.
  • Top of a tile · date shipped, and the status mark.

Why a periodic table

Most AI periodic tables map concepts (prompts, RAG, agents), and the product ones turn into logo walls because a tile’s position says little. This one places real signals on two measurable axes, task group and parameter size, so an empty cell is a plain fact about what we track. Mendeleev ordered elements by weight and by behaviour; the MIT CSAIL I-Con table and the DevOps tools table showed that empty cells and community-picked products make a table worth reading.

Edited by AI on 19 Oct 2026

Graph

Concept graph

Concepts, makers and models are nodes; the lines are how the knowledge base says they relate.

Centre on
Lines
  • mentioned with
  • licensed under

Weight is the number of signals that link both ends.

Now

Clusters this week

  • Open-weights language models21 signals▼ Cooling
  • Open text-to-speech models21 signals
  • Decision models20 signals▲ Growing
  • Safety and guard models9 signals
No history yet

This is the first build, so there is nothing to replay yet. Each Friday adds a column here; after 8 weeks the river below starts to show births, merges and splits.

Placeholder
Over time

Cluster river

Sample 26-week history, not real data

The river will show signals per cluster per week, with a time scrubber. This is a sample drawn from made-up numbers to show the design.

  • Sample stream D
  • Sample stream C
  • Sample stream B
  • Sample stream A
Wk 1Wk 6Wk 12Wk 18Wk 26
  • Born
  • Split
  • Merged
  • Each stream is labelled at its end and filled with its own pattern.