Your Claude quietly
became Fable 5.
Until last week, the default for both claude.ai and the API was Opus 4.8. It has now switched to Claude Fable 5—open the UI and anyone gets it today, or on the API just specify one line for the model ID. The first publicly available Mythos-class model: here's a visual breakdown of what changed and what to watch for in production.
From "what is this?"
to the official default
Around the time of release, an unfamiliar model ID, claude-fable-5, suddenly appeared in Anthropic's SDK, and on X people were asking "what model is this?" The answer: the next flagship. Released on June 9, 2026, "Claude Fable 5" became available in the Claude UI and API as a generally available model—the first publicly available Mythos-class model.
The key point is that the switch happened quietly, yet across the board. Open claude.ai and, without changing any settings, the flagship is Fable 5. On the API too, just specify claude-fable-5 as the model ID and you can use it the same day. The Anthropic Python SDK already supports both models in v0.108.0 / v0.109.0, so bumping the library is all the preparation you need.
| Until last week (Opus 4.8) | From today (Fable 5) |
|---|---|
| UI / API default is Opus 4.8 | The UI flagship is Fable 5 |
| claude-fable-5 is an unidentified ID | Officially generally available, specifiable the same day |
| The Mythos class is not public | First publicly available Mythos-class model |
| Opus 4.8 is the ceiling for long, complex tasks | Said to surpass the current generation in the same areas |
The new default has
quietly arrived right beside you.
5% quietly falls
back to Opus 4.8
Fable 5 has a built-in safety mechanism. For certain inputs, the response quietly falls back to the previous generation by design.
Fable 5 is said to surpass the current Opus 4.8 generation on long and complex tasks, and is designed so that a safety classifier quietly falls back to Opus 4.8 about 5% of the time. In other words, even for the same prompt, the model that returns the response is not always Fable 5.
This is not a quality degradation but a mechanism to err on the side of safety. That said, when you collect logs or evaluations in production, it's safer to assume that "which model actually answered" is not always fixed. If you observe variability in outputs, this fallback is the first thing to suspect.
Who should switch,
and when
How much you benefit depends on how you use it. Here's "what to do today" organized by role.
People building products on the API
Specify claude-fable-5 in your existing calls, or ride the default, and the capabilities are available today. Start with a small A/B test first.
Heavy long-form and coding users
If you use long-form summarization, coding, and complex instructions daily, the switch is worth a try. These are areas where the difference is easy to feel.
People who only use it occasionally
If you only touch it a few times a week, the difference may be within the margin of error. You can keep using the default with no problem.
"Mythos" has
come into the open
What makes this update symbolic is that a Mythos-class model, previously kept private, has reached the general public for the first time. Each time the flagship turns over a generation, the model we use every day is quietly swapped out without us doing anything. We've entered an era where the identity of "the usual Claude" is being updated before we notice.
That is exactly why, in production, the stance of pinning the model explicitly and observing changes deliberately pays off. Ride the latest capabilities right away, or verify first and then adopt—it's realistic to see this as a widening of your options. Start small, watch the behavior including the fallback, and then roll it out to production.