The obvious problem with naming your best model Opus is that you eventually have to build a better one.

A magnum opus is supposed to be the work someone is remembered for. Anthropic’s original Claude hierarchy—Haiku, Sonnet, Opus—was memorable and reasonably intuitive. A haiku is short, a sonnet is more substantial, and an opus is a major work (in terms of impact, but that’s often correlated with length). The metaphor was not mathematically precise, but it did not need to be. Many could see the three names and understand the intended order.

It was also versioned. Claude 3 Opus could become Claude 3.5 Opus, then Claude 4 Opus, without disturbing the hierarchy. The model family and the generation were separate concepts.

But Opus left no obvious room above it. Anthropic eventually introduced Mythos as a new class above Opus, with Fable as a safeguarded model built from the same underlying system. The etymology is defensible: fable and mythos both refer to stories, and Anthropic explains the relationship between them. As a product hierarchy, though, Haiku → Sonnet → Opus → Fable is no longer something a person can infer. The metaphor has moved from the length of the writing, to the importance of the work, to the nature of the story itself.

Anthropic outgrew its own naming scheme.

What’s in a name? For a product company, quite a lot. Names can feel cosmetic next to the engineering work, but they are part of the interface. A name tells a customer which product is newer, which is more capable, which is faster, and which one the company expects them to buy. When the name stops carrying that information, every pricing page and product announcement has to add it back.

Naming systems tend to hold up when they keep a few different ideas separate: product family, capability tier, generation, and specialization.

Canon’s older DSLR camera lineup handled this reasonably well, for a slow-moving product. The leading number broadly indicated the product’s place in the range, in which the lowest number indicated the most premium offering, while “Mark II,” “Mark III,” and “Mark IV” suffixes indicated generations within it. The gaps between the models—1D, 5D, and 7D—also left room to introduce the 6D without renaming everything around it. Suffixes such as “s” could identify a specialized variant. The complete catalog was not perfectly orderly, but a name like “5D Mark IV” carried a surprising amount of information. For a physical product line, which tends to evolve and branch more slowly than software, that convention provided enough headroom.

Venture capital funding rounds use an even more boring system: Series A, Series B, Series C. The letters make no claim about what a company has become. They only establish sequence (which often, but not always, implies higher valuations in later rounds). That has been enough to survive companies remaining private far longer than the convention was designed for. If a startup ever reaches Series Z, spreadsheets have already given us a perfectly serviceable answer: Series AA.

Apple is at its clearest when it uses ordinary words to describe an actual tradeoff: Air for portability and Pro for capability. Its catalog becomes less precise with Plus and Max, and harder still when those modifiers start stacking as Pro Max, but the direction remains somewhat legible. Although its naming has become more convoluted over time, a customer does not need a literature degree to guess whether a Pro model sits above the baseline one.

AI companies have made this problem unusually visible because their product lines change (and get shut down) so quickly. OpenAI had already accumulated GPT, Codex, and models such as o1 and o3—names that mixed product, model family, generation, and capability. With GPT-5.6, it introduced Sol, Terra, and Luna as durable capability tiers, while keeping the version number for the model generation. Structurally, that is a good decision.

The names themselves are less convincing. Sol is the most capable model, Terra is the middle tier, and Luna is the fastest and least expensive. The apparent scale is mass—Moon, Earth, Sun—although OpenAI’s announcement does not explain it. I am particularly interested in space and still first read it incorrectly as the distance from Earth: Terra, then Luna, then Sol. The two interpretations produce different orderings, and neither provides much room for an obvious fourth tier. What comes after the Sun? Another star is easy to name, but its relative mass is not common knowledge. What comes below the Moon? And more broadly, how many users will infer any of this from the names alone?

There is a real tradeoff here. Standard, Pro, and Max feel generic. A bespoke metaphor feels ownable and personal, and with three products it can make a lineup feel unusually coherent. Furthermore, it makes for a great press release. But the cleverness creates a debt that comes due when the roadmap changes. A naming scheme has not been future-proofed simply because it can technically continue; it has to remain understandable as it does.

The systems that age best are usually a little boring. They have an obvious direction, leave room on both sides, and distinguish a new generation from a new tier. Most importantly, they do not require users to remember the internal logic behind them.

Naming is hard because it compresses positioning, capability, and brand into a few words. Future-proof naming is harder because it also has to account for products that do not exist yet.

Successful companies can bet that their best product today will not be their best product forever. It helps not to name it as though it is.