Every Anthropic model name, ranked
From Haiku to Fable to Cinematic Universe. Anthropic's naming scheme has gotten weird. Every model name from the Sam Wilkinson HN post ranked by how well it describes the model.
TL;DR: Anthropic’s model names started elegant (Haiku, Sonnet, Opus) and got weirder (Mythos, Fable). Then Sam Wilkinson’s satirical HN post extrapolated the naming scheme into absurd territory, Terms of Service, Cinematic Universe, Saga (Unabridged), and it took off. Here’s every name ranked.
On June 9, Sam Wilkinson published “Anthropic’s Model Naming, Extrapolated”: a one-minute-read chart that extends Anthropic’s naming scheme to its logical, absurd conclusion. It hit 277 points on Hacker News with 77 comments. Because of course it did. Developers love a good naming-scheme roast.
Here’s every name, real and imagined, ranked by how well it describes the model.
Key takeaways:
- The real naming scheme (Haiku → Sonnet → Opus) is genuinely good. Each name tells you something about cost and capability. Haiku is quick and cheap. Sonnet is the balanced workhorse. Opus is deep and expensive.
- Mythos and Fable broke the pattern. Moving from poetic forms to narrative genres was a signal that Anthropic’s product line had expanded beyond the original three-tier framework. Mythos = research preview. Fable = mythos with safety. The names work but the logic is less intuitive.
- Sam Wilkinson’s satirical table is funny because it fits. Terms of Service, Cinematic Universe, Overwhelmingly Large Narrative Unit: these jokes work because Anthropic’s naming trajectory makes them feel plausible.
- The naming scheme reveals strategy. The shift from poems (Haiku, Sonnet, Opus) to narrative objects (Mythos, Fable) mirrors Anthropic’s move from API provider to platform company with multiple product lines.
What are the real Anthropic model names ranked?
1. Haiku
What it is: Small, fast, cheap. The pure distillation model.
The name works because: A haiku is short, structured, and complete. That’s exactly what this model does: quick tasks with low latency. The name tells you everything: small bill, fast output, don’t ask it to write a novel.
Does it fit? Perfectly.
2. Opus
What it is: The flagship. Long, deep, expensive. The model you reach for when you need the best answer.
The name works because: An opus is a large, significant creative work. It’s what you call a composer’s masterpiece, not a sketch. The name signals: this is the big one, use it when it matters.
Does it fit? Perfectly.
3. Sonnet
What it is: The workhorse. Structured output, balanced cost, reliable for most tasks.
The name works because: A sonnet sits between a haiku and an opus: more structured and substantial than a haiku, but not a full epic. It’s the middle child that gets most of the work done. The name is the weakest of the three originals only because there’s nothing wrong with it.
Does it fit? Yes, but it’s the least distinctive of the three.
4. Mythos
What it is: Research preview of Anthropic’s next-generation architecture. Near-frontier capabilities, restricted access. The model that powers Glasswing partnerships.
The name works because: A mythos is a foundational story: the underlying narrative that shapes everything else. This model is the foundation for Fable. It’s appropriate for something most developers can’t access but hear about constantly.
Does it fit? Yes, but the naming logic has shifted. We’re no longer in “poetic forms” territory. Mythos is a genre, not a form.
5. Fable
What it is: Mythos with safety guardrails. The same underlying model, released to the public with automated classifiers that detect and block high-risk queries.
The name works because: A fable is a story with a moral: it teaches you something. Fable, the model, is Mythos made safe. The safety layer is the moral. The name flip from Mythos (the raw foundational story) to Fable (the moralized version) is honestly clever.
Does it fit? Yes, but you need to know the context. New users won’t get the Mythos → Fable connection without reading the docs.
What are the satirical Anthropic model names?
The post is structured as a table extrapolating Anthropic’s naming scheme. Here are the highlights, ranked:
1. Terms of Service
Joke: “No liability for answers or the consequences thereof.”
Why it’s the best: Because every developer has had this exact thought while building an AI product. It’s not a model: it’s a legal document. The fact that it’s in the same list as Haiku and Sonnet makes it land perfectly.
2. Overwhelmingly Large Narrative Unit
Joke: “Requires viewing a ‘previously on’ segment prior to use.”
Why it works: This is what happens when you stretch the naming scheme past its breaking point. The absurd formality is the joke. It also accurately describes what Claude with a million-token context window feels like.
3. Saga (Unabridged)
Joke: “Includes answers to unrelated questions.”
Why it works: Anyone who has used a large model for a complex task has experienced this. The model gives you everything, including things you didn’t ask for. It’s a self-own and it’s accurate.
4. Cinematic Universe
Joke: “Multiple Sagas with a Lore dispatch layer.”
Why it works: This is a genuinely good description of multi-agent architectures. The fact that it started as a joke about model naming makes it funnier. Expect to see this term in actual architecture docs within 6 months.
5. Marginalia
Joke: “Provides unprompted commentary on your code.”
Why it works: If you’ve used an AI coding assistant that keeps adding suggestions you didn’t ask for, you’ve used Marginalia. It’s the first model name that’s a UX complaint.
6. Diatribe
Joke: “Sonnet, but angry.”
Why it works: The simplest joke in the list and one of the best. A sonnet is structured and balanced. A diatribe is the same form but with strong opinions. That’s a real product distinction if you think about it.
7. Abstract
Joke: “Summarizes reasoning it hasn’t done.”
Why it works: Painfully accurate for every LLM that confidently explains its reasoning for a wrong answer. The name is a three-word paper on hallucination.
8. Treatise
Joke: “Opus, but citation is left as an exercise for the reader.”
Why it works: References the LLM hallucination problem again. A treatise without citations isn’t research: it’s what every chat model does.
9. Lore
Joke: “Interpretation requires a wiki.”
Why it works: For context-dependent models where you need a separate documentation site to understand the output. Also a solid description of Claude with custom instructions and long context.
10. Aphorism
Joke: “One sentence, but it always feels right.”
Why it works: This describes the Haiku tier experience when you ask for a summary and get a single, perfect sentence. It’s a nice model name.
What the naming evolution tells us
The shift from poetic forms (Haiku, Sonnet, Opus) to narrative genres (Mythos, Fable) isn’t random. It maps to Anthropic’s product strategy, which I covered in my Claude Fable 5 first look and the one week in roundup:
- Stage 1 (poems): Three models, one API. The names are a clever mnemonic for developers choosing between speed, balance, and capability.
- Stage 2 (narratives): Multiple product lines. Mythos (research), Fable (safe mythos), Claude for Work (enterprise), Claude Code (developer tool). The simple three-tier naming couldn’t contain the product surface anymore.
Wilkinson’s extrapolation works because it follows this trajectory to its logical extreme. If every product line needs a literary genre name, eventually you’re shipping Terms of Service and Cinematic Universe.
The real question: what comes after Fable? The HN thread had theories. My bet is on “Chronicle”: a model improved for long-context recall and audit trails. But at this point, I wouldn’t rule out “Epilogue.”
FAQ
What are all the real Anthropic model names? The real model tiers are Haiku (small/fast/cheap), Sonnet (workhorse), Opus (flagship), Mythos (research preview, restricted access), and Fable (Mythos-class model with safety guardrails, publicly available). Version numbers like 4.6, 4.7, 4.8 track improvements within each tier.
What’s the Sam Wilkinson HN post everyone is sharing? Sam Wilkinson published ‘Anthropic’s Model Naming, Extrapolated’ on June 9, 2026. It creates a satirical table of made-up model names : Aphorism, Marginalia, Abstract, Diatribe, Treatise, Saga, Lore, Cinematic Universe, Terms of Service : each with a one-line joke. It got 277 points on Hacker News.
Why did Anthropic choose Haiku, Sonnet, and Opus? Anthropic uses literary forms as naming tiers : Haiku (short/concise) for the cheapest model, Sonnet (structured/balanced) for the workhorse, Opus (large/complex) for the flagship. It’s a clever mnemonic: the name tells you about cost and capability.
What is Claude Mythos? Claude Mythos was a research preview model released in April 2026 with near-frontier capabilities, available only to approved researchers and Glasswing partners. It became the basis for Claude Fable 5, which makes Mythos-level capabilities available to general users with additional safety guardrails.
Related Posts
- Claude Fable 5 first look: benchmarks and developer reactions
- Claude Fable 5 one week in
- Best AI coding agents 2026
Anthropic’s official model documentation lists all Claude models and their capabilities. A public timeline of Anthropic model releases tracks every Claude model launch date and naming pattern.
Anthropic’s model documentation lists all Claude models and their capabilities.
This article was published on Agentic Up (https://agenticup.dev): practical guides for developers and founders building with AI agents. Reach me at hello@agenticup.dev.
