Which Claude model you actually get now depends on your plan

For about six weeks, the honest answer to "which Claude model should I use" was wait, it'll change again. It changed four times. That stretch is over, and what replaced it is stable — but it comes with a twist that no amount of reading about model quality will tell you.

You no longer just pick a model. You pick a plan, and the plan picks your ceiling.

Two things landed in the same week. On July 20, Anthropic settled Fable 5's availability permanently. On July 24, Claude Opus 5 arrived. Together they mean the model at the top of your picker may be different from the one at the top of your colleague's, and the difference is what you each pay per month rather than anything about your work.

Here is the current state, then what it actually means for a consulting practice — which is a different question, and the one worth your time.

The state, as of today

Current as of July 24, 2026. Check Anthropic's own plan page before acting on this — it is the authority, and this is the fifth time the answer has changed since June.

Your plan	Your top model	Fable 5
Free	Sonnet-tier	Not available
Pro	Opus 5 — the strongest model on Pro	Not included; reachable with pay-as-you-go usage credits
Max	Fable 5 — with Opus 5 as the default	Standard, up to half your weekly usage; usage credits past that

Team and Enterprise follow the same split by seat type — premium seats behave like Max, standard seats like Pro. Ask whoever owns your workspace.

Anthropic describes Opus 5 as "the new default model on Claude Max, and the strongest model on Claude Pro," and says it "comes close to the frontier intelligence of Claude Fable 5 at half the price." It also carries an adjustable effort setting, which turns out to matter more day to day than the plan question does.

That's the whole factual picture. Everything below is judgment.

The mistake this invites

The obvious reading is: the good model now costs more, so pay more.

That reading is wrong for most solo consultants, and it's worth being concrete about why. Reaching Fable 5 as a standard part of your plan means Max, which starts at $100 a month against Pro's $20. That's a real difference — but the relevant comparison isn't $20 versus $100. It's $80 a month against what those eighty dollars actually change about your work.

At a consulting rate of $150 an hour, $80 is about thirty-two minutes. So the question becomes: does having Fable 5 rather than Opus 5 at the top of your picker save you more than half an hour a month?

For a practice whose AI work is proposals, client updates, case studies, meeting debriefs, and scoping — which is most practices — I don't think it does. Not because Fable 5 isn't better; Anthropic's own framing is that Opus 5 comes close to it, not that it matches it. But "close to the frontier" is already well past the difficulty of drafting a Friday update from a week of messy notes. The binding constraint on that work was never model capability. It's whether you handed over a complete brief, which is a you problem and stays a you problem on every tier.

Where the ceiling genuinely binds

There is real work where the top tier earns its money, and it has a shape worth recognising:

Long, single-handoff assembly. A forty-page RFP you need absorbed and answered. A case study assembled from a project folder with a dozen sources. Seven stakeholder interviews compressed into one diagnostic. These are the tasks where the model has to hold the whole job coherent across many steps, and where the middle of a long run is exactly where weaker models drift. This is the long-leash end of the routing rule, and it's the one row in that table where your plan decides for you.

Work you can't easily check. The failure mode that costs you is the one that survives your review because the output looks finished. A proposal with a plausible-but-wrong price. A case study citing a detail the client never said. The more finished a long-run draft looks, the more you want the strongest model that made it — and the harder you should read it regardless.

Volume at the top of the range. If you're running several long handoffs a week rather than one a month, per-use credits stop being the cheaper path and a plan that includes the tier starts making sense on arithmetic alone.

If none of those describe your month, the plan-included model is not a compromise. It's the right tool, and the eighty dollars is better spent on almost anything else.

The lever that actually changed your week

Buried under the plan news is the thing more likely to affect your work tomorrow: Opus 5 has an adjustable effort setting — you can dial how much the model does before it answers.

This maps onto the routing rule so cleanly it's almost suspicious. Short leash, low stakes, you're reading every word anyway: leave effort low and enjoy the speed. Long leash, expensive wrongness, a draft you'd hate to get subtly wrong: turn it up and let the model verify itself before you see it.

The practical effect is that the gap between "the strong default" and "the top tier" narrowed for most consulting work. A well-briefed Opus 5 run at high effort covers a lot of the deliverable work that used to mean reaching for the tier above — which is precisely why the plan question matters less than the headlines suggest. It deserves its own treatment, and it'll get one shortly.

What to do this week

Open your model picker and read it. Not an article — your own picker. It reflects your account, and after this month it's the only source of truth about what you can actually reach. Whatever sits at the top is what every routing recommendation means by "your top tier."

Run one honest comparison before changing anything. Take a deliverable you produce regularly. Same inputs, same brief, run it on your current top model at high effort. Read the output cold and ask whether a stronger model would have changed the outcome or just the adjectives. That's evidence from your practice, which beats anyone's benchmark table — including the one above.

If you want a ready-made deliverable to run it on rather than improvising one, our free Friday Update Brief is a complete handoff — the brief, a sample week of messy notes, and the reference output to compare against. It costs an email address, and it makes the comparison fair because both runs work from identical inputs. That's the cheapest honest test of whether your ceiling is actually limiting you.

Don't reorganise your practice around a plan in its first month. The structure just settled after four changes in six weeks; Anthropic has committed to this one, and it's reasonable to believe them. But the consultants who got hurt this summer were the ones who rewired fastest around whatever had just been announced. Try things eagerly, rewire slowly.

The part that didn't change

Every model in that table reads the same context files, follows the same standards, and runs the same briefs. The setup that makes AI useful for your practice specifically — who you serve, what you sell and at what floor, how you sound, what's true of each client right now — is model-independent and plan-independent by construction. It was the thing worth building in June, it survived four changes of terms since, and it will survive whatever lands in September. We've made that argument at length, and this summer tested it about as thoroughly as anyone could ask.

If you'd rather start from finished files than a blank page, the Claude Project Kit ships that standing setup — the instructions block and the four practice files — and it works identically on whichever model your plan puts at the top of your picker. That's not a hedge. It's the whole point.

The model at the top of your picker is now a billing decision. What you hand it is still the job.