Claude Fable 5, Day Two: More Hours, Less Magic

On day one, Claude Fable 5 felt like a product owner — it dropped the data dump, reframed my work around outcomes, and reached the same pivot-or-quit conclusion I'd reached myself. On day two, across far more hours and far more projects, it felt like a very good analyst again. Still better than Claude Opus 4.8, just not by the margin it seemed at first. I want to be honest about that, because the change I'm describing might be in the model — or it might be entirely in me.

This is a follow-up to my first-hours impressions of Claude Fable 5, written after Anthropic shipped it on June 9, 2026 as its first public Mythos-class model. Yesterday, June 10, was the second day it appeared in Claude Desktop — and again, like that first Tuesday, it lasted only a few hours. But this time I got more done with it before it vanished, which is exactly why my read on it shifted.

What I Actually Did With It on Day Two

The first day was a few intense hours and one conversation that floored me. Day two was wider. I ran Fable 5 across a real spread of work: brainstorming entirely new business models, pressure-testing ideas, and the unglamorous middle of my week — improving projects I already run, tightening copy, reworking structure, thinking through priorities. It was the first time I'd used the model the way I use Claude every day, rather than as a shiny new toy I was poking at.

And that breadth is the point. A single dazzling answer is easy to over-weight. Across a dozen different tasks, you start to see the steady-state behaviour rather than the highlight reel — and the steady state was good, just less electric than the trailer promised.

The Magic Dimmed — and I Can't Fully Explain Why

Here's the honest part. On day one I wrote that Fable 5 announced itself in a single answer, that it managed instead of merely answering, that it thought like a product owner rather than an analyst. On day two I kept waiting for that to happen again, and it mostly didn't. The answers were sharp and useful. They just weren't surprising. It felt more comedido — more restrained, more careful, more inside the lines — than the model that had genuinely excited me twenty-four hours earlier.

Day one, Fable 5 told me what to do. Day two, it told me what I'd asked. Both are useful. Only one made me sit up.

Did Anthropic change the model? I doubt it

My first instinct was the obvious one: did they quietly dial it back between Tuesday and Wednesday? I want to be clear — I have no evidence of that, and I genuinely doubt it. Companies don't usually swap the weights of a flagship model overnight without saying so, and "it felt different" is the weakest kind of evidence there is. I'd love for Anthropic to confirm one way or the other, but absent that, the responsible read is that the model didn't change. I did.

The more likely explanations are unromantic:

• Novelty and expectation. The first encounter set an impossibly high bar. By day two I was no longer comparing Fable 5 to Opus 4.8 — I was comparing it to my own memory of day-one Fable 5, which had already been polished by excitement.
• A different task mix. The day-one moment was a hard, open strategic call — exactly where a decisive model shines. Day two was mostly execution and refinement, where there's less room to dazzle.
• Its own admitted variance. Fable 5 told me itself that it takes bigger swings and is therefore more likely to miss. Higher ceiling, higher variance. A model like that will feel uneven across a long session almost by design — brilliant in one thread, merely competent in the next.

None of those require Anthropic to have touched anything. All of them produce exactly the feeling I had.

Still Better Than Opus 4.8 — Just Not "Much Better"

To keep myself honest, I held the comparison where it belongs: not against my memory, but against the model I use most. Measured against Claude Opus 4.8 — which itself was only a quiet, careful step up from 4.7 — Fable 5 was still clearly the stronger model on day two. Tighter reasoning, better instincts about what mattered, more willing to take a position.

The recalibration is only about size. On day one the gap felt like a different class of machine. On day two it felt like a real, worthwhile upgrade — the kind you'd happily pay for if the price and the access made sense — but an upgrade, not a leap. That's still a good outcome. It's just a more sober one, and I'd rather report the sober version than ride the day-one high into a verdict I can't back up.

Still Waiting for Fable 5 in Claude Code

The thing I most want to test, I still can't. I'm on Claude Max 5×, and Fable 5 has not yet shown up for me inside Claude Code — only in the chat apps, and only in those short windows. For my work that's the gap that matters: chat is where you judge a model's thinking, but Claude Code is where you find out whether the thinking survives contact with a real, multi-file codebase. A model that's decisive in a conversation and a model that can be trusted to drive an agentic coding session are not the same claim, and I've learned to budget real debugging time before I trust either.

So I'm waiting for a usage window to open in Claude Code, and I'm impatient about it. The day-one chat impression was enough to make me want the coding version badly. Until then, anything I'd say about Fable 5 as a coding model would be a guess, and I'd rather not.

Gone Again Today — and That's Probably the Pattern

As of this morning, June 11, Fable 5 is absent from every Claude app I have. That tracks with how the first two days went: it surfaces for a few hours, then disappears. The likeliest reason is the same one I landed on after day one — it's expensive to run ($10 / $50 per million tokens, double Opus 4.8) and metered tightly, so even on a high-tier plan you burn the allotment fast, and the model itself appears to be rationed while it's still being stabilised. Remember the access terms, too: Fable 5 is free on Pro, Max, Team and seat-based Enterprise plans only through June 22, then it moves to usage credits.

My working assumption is that intermittent, here-and-gone availability will be the norm until the model is more stable and consolidated. Annoying, but understandable for a frontier release this fresh.

The Bottom Line

Day two cooled my day-one excitement without erasing it. Claude Fable 5 is still the most capable model I've touched, and still ahead of Opus 4.8 — but across many real hours of work it read as more measured than the product owner that wowed me on Tuesday, and I'm honest enough to admit that the difference might be in my own expectations rather than in the weights. I doubt Anthropic changed anything; I'd just like them to say so. What I'm sure of is that I haven't seen the version I most care about yet — Fable 5 in Claude Code — and the moment a usage window opens, that's the test I'll run. The full verdict waits until I can keep the model long enough to earn one.

FAQ

Did Anthropic weaken Claude Fable 5 after launch?

There's no evidence of that, and I don't believe it did. My day-two impression that Fable 5 felt more restrained is subjective; "it felt different" is weak evidence, and companies rarely change a flagship model's weights overnight without announcing it. The more likely causes are the novelty wearing off, a different mix of tasks, and the higher output variance the model openly describes in itself. I'd welcome an official clarification from Anthropic, but the responsible assumption is that the model didn't change.

Is Claude Fable 5 still better than Claude Opus 4.8?

Yes. In my second day of use, Fable 5 stayed clearly stronger than Claude Opus 4.8 — tighter reasoning and better instincts about what matters. What changed was the perceived size of the gap: on day one it felt like a different class of model, on day two like a real but ordinary-sized upgrade.

Can you use Claude Fable 5 in Claude Code?

Not in my experience yet. I'm on Claude Max 5× and Fable 5 has only appeared for me in the Claude chat apps so far, not inside Claude Code, and only during short windows. I'm waiting for a usage window to open there, because agentic coding is the test I most want to run on it.

Why does Claude Fable 5 keep disappearing from the apps?

Fable 5 has surfaced for only a few hours at a time on each of its first days. The likeliest explanation is that it's expensive to run — $10 / $50 per million tokens, double Opus 4.8 — and metered tightly, so usage caps are hit quickly, while the model is still being stabilised. It's free on Pro, Max, Team and seat-based Enterprise plans through June 22, 2026, then moves to usage credits.

Should I trust a second-day impression of an AI model?

Treat it as a data point, not a verdict. A first impression can be inflated by novelty and a second can over-correct; neither is a substitute for sustained, varied use. The honest takeaway from day two is calibration: Fable 5 is better than Opus 4.8, the day-one "leap" was probably part real upgrade and part excitement, and the real test — long sessions and Claude Code — is still to come.