Meta has decided that its AI models will no longer be named after fruit, Business Insider reported on November 9. The reasoning is that as the models get better, the company wants names that read like products rather than codenames.
The decision is small and the reasoning is not. A codename is a cheap object: it can be changed without a customer noticing, it carries no promise about a release date, and it lets a lab ship something, learn from it and rename the next version without anyone treating the first as a product.
That is why labs use them. Fruit was a good choice for exactly that reason, memorable internally, meaningless externally, and impossible to confuse with a commitment. A product name is the opposite instrument.
A product name implies a customer, a support path, a version history and a comparison with whatever a competitor has on the shelf. It also dates. The moment a model is called something a buyer can ask for, the company has to say what it costs, what it replaces and when the next one arrives.
Meta is giving up the option value of the codename in exchange for the selling power of the product name. The trade only makes sense if the company believes the selling is about to matter more than the shipping. It is a bet on timing, made with the cheapest asset a lab owns.
There is a cost inside the building, and it is larger than it looks. Teams talk in the shorthand they were given, and documentation, dashboards, evaluation sets and the slide a researcher shows on a Tuesday all carry the old names. Every one of them has to be migrated or explained.
The migration is tedious and the explanation is worse, because a rename invites the question of what changed, and the answer is nothing. The leak channel also changes. A fruit name in a job posting or a benchmark submission tells an attentive reader very little beyond the fact that something exists and roughly where it sits in a sequence.
A product name tells the market what the thing is for, who it is aimed at and where it sits against a rival’s line. That is more information, released earlier, and it is information a competitor can price against. Meta is trading a quiet channel for a loud one because the loud one sells.
The move also fits a pattern that repeats across the industry. Research organizations name things for internal clarity. Commercial organizations name things for the customer’s benefit, which usually means a word that describes a job rather than an ingredient.
The shift from one habit to the other tends to arrive before the revenue does, because naming is cheap and it signals intent to the sales channel, to partners and to the analysts who build models of the business. A company that renames its models is telling those audiences that the next phase is about distribution. That is a claim a company makes when it believes the product is ready to be sold rather than evaluated.
The risk is that a product name arrives before the product does. Call a model something a buyer can order and the buyer will expect what a product implies: reliability, a support contract, a roadmap, a price that does not move every quarter. If the capability is still improving on a schedule the company cannot publish, the name makes a promise the release cadence has to keep.
The counterargument is that none of this touches the model. Renaming does not change a benchmark, a latency figure or a safety evaluation, and the company said as much. The change is in external communication, which is the layer where adoption is won and lost, and where a good model with a bad name loses to a worse model with a clear one.
None of this makes the old system wrong, and it is worth saying why it lasted. A codename buys time: it lets a team test a capability with users before the company decides whether the capability is a product, and it keeps a failed experiment from becoming a brand problem. Meta is not abandoning that discipline so much as declaring it finished for this generation of models.
The fruit was useful while the question was whether the thing worked. The question now is who buys it, and that is answered with a name a procurement officer can write on a form. What the naming decision reveals is a view about timing.
Fruit names belong to a period when the models were experiments with release dates set by researchers. Product names belong to a period when the release date is a commitment to a customer. The test is not the name: it is whether the next launch arrives with a price, a support path and a date that holds. If those appear, the fruit bowl was retired at the right moment; if only the name changes, the company has spent its cheapest signal on nothing.
