2026-11-13 · ← News
Nine Model Names and One Needless Puzzle
Ethan Mollick packed nine AI model names into one post to show how hard it has become to explain which model corresponds to which. For buyers and teams, that confusion is practical: the names no longer help with selection.
The image could not be loaded.
In a November 13 post, Ethan Mollick arranged the names hydrogen, sonnet, luna, spark, fable, wonder, astra, carbon and argon into a single sentence about which model resembles which. The joke works because readers must decode the vocabulary before they can discuss model capabilities.
Nine names obscured a simple comparison
Mollick does not claim to be publishing a benchmark or a new product map. He is caricaturing how people talk about models: one family uses literary labels, another reaches for space and another for chemical elements. The post compares hydrogen with sonnet, luna and spark, then fable with wonder, astra and carbon, while noting that it used to resemble argon.
The only firm fact is the text of the post. Its equivalences should not be read as measured results. The user facing point is clearer: without a private alias table, even an informed reader can get lost before sending the first prompt.
Model selection is turning into translation between catalogs
For a team buying an API or configuring internal tools, a name is not decoration. It should help identify generation, size, price, speed and support lifetime. When every vendor builds a poetic taxonomy, users must reconstruct those details from documentation, pricing pages and benchmarks.
The practical cost appears during change approval. A proposal to move from argon to carbon does not reveal whether carbon is cheaper, newer or merely another mode of the same product. Without an exact version identifier, an ordinary operational decision becomes translation work.
A sharp joke is not a map of real products
The post is commentary, not procurement evidence. It provides no prices, context windows, availability details or eval results. Its nine names cannot support a ranking or recommendation.
It would also be too easy to blame branding alone. Models differ by interface, region, plan and release date, and one short label cannot carry all of that. The problem begins when a vendor fails to pair the brand with a stable technical identifier and a clear change history.
Stable identifiers and legible pricing will settle the issue
A useful signal will be documentation that links each commercial name to an immutable ID, release date, price and retirement date. More important still will be the ability to pin a production version and receive notice before it changes.
If teams keep explaining models with sentences such as “fable used to be argon,” vendors have shifted the cost of their nomenclature onto customers. Genuine simplification will be visible when a decision can pass review without a private glossary.
Lilith's verdict
Today’s model catalog resembles a station board where every operator gives the same service a different name. The passenger needs a train number and departure time, not another nickname.