The Sonnet 4.8 that never was: a leaked codename isn't a release

For months the community has been asking "when is Claude Sonnet 4.8 coming?" As of June 19, 2026, the clear answer is: there is no such model. No official announcement, no API model id, no benchmarks, no date. Anthropic's current Sonnet is still 4.6.

So why was everyone so sure "4.8 is coming"? The answer lives in one of the most confusing parts of modern AI watching: leaked codenames.

What the leak said — and didn't say

The story traces back to the Claude Code source that leaked on March 31, 2026. What communities quickly found, beyond unreleased features, were explicit references to future model versions and a complete internal codename system. The string "Sonnet 4.8" appeared in that forbidden-strings list.

The conclusion drawn from it was wrong: a version number showing up in code does not mean that name will ship publicly. The string only confirms internal intent — a planned slot, a codename, a possibility. Between intent and a shipped product sits the whole distance of a product team, safety testing, and strategy.

Why intent didn't become a product

Here's the interesting part: while the Sonnet line stayed stuck at 4.6, the "next leap" went elsewhere. Anthropic made the expected upgrade not as Sonnet 4.8 but directly in the Mythos-class tier. The Capybara/Mythos line went public as Claude Fable 5.

So the "4.8" intent we saw in the leaked list wasn't wasted; it just shipped under a different name and a different positioning. Meanwhile what actually launched was Opus 4.8 — so the "4.8" number did arrive, but attached to Opus, not Sonnet. The name-matching tangled up in the heads of people reading the leak.

Why I care about this

This isn't just trivia; as a developer building a roadmap, this is exactly the trap that's easy to fall into. Someone planning "I'll migrate when Sonnet 4.8 ships" has tied themselves to a date that doesn't exist.

Leak-based news has three quiet problems:

• Name ≠ product: a codename is a slot, not a commitment.

• Number ≠ position: "4.8" can belong to one or several products, not a family.

• Date ≠ leak: when it ships isn't written in code; that's a strategic call.

For a solo builder the practical rule is simple: build your plan on top of models that have shipped and have an API id. Not on the expectation model.

Limits / concerns

Ignoring leaks entirely isn't right either. That list carried real information — it correctly showed Anthropic's internal codename system and general direction. The issue isn't the value of the information, but the certainty people loaded onto it.

By the same token, I can't say "Sonnet 4.8 will never ship." Anthropic could use that name tomorrow. All I'm saying is: as of June 19 it doesn't exist, and basing a roadmap on something that doesn't exist produces fragility.

A note from me

Part of my reason for writing this is a reminder to myself. I'm someone who reads leak headlines with excitement too; "this model is coming" stories give a dopamine hit. But I've learned to leave that excitement at the door when I make product decisions.

In AI watching, the healthiest stance is probably this: follow the rumor with curiosity, but build your architecture only on what actually works in your hands. The next model's shiny benchmark doesn't run today's product; the shipped model does.