Which model should I write with?
Open the model menu in most AI tools and you get a list of codenames: version numbers and suffixes that mean a great deal to the companies shipping them and very little to anyone mid-scene. Inkwell's picker asks a more useful question: how much model does this pass need? Every account gets a three-tier ladder — fast, balanced, frontier — and each tier resolves to a specific, named model on whichever provider serves you.
The ladder, in writing terms
The fast tier powers everything that has to feel instant. Suggestions while you type always run here, whatever you have picked, because latency matters more than depth when you are mid-sentence. It is also the right choice for quick mechanical passes where you want speed and volume over subtlety: a formatting sweep, a first look at a scene you just drafted.
The balanced tier is the everyday workhorse. A dialogue pass on a scene, a punch-up on an act, the skill runs that make up most of a working session. On Pro, this is the default; if you never open the picker, this is what serves you.
The frontier tier is the most capable model on the ladder, and the one to reach for when a pass has to hold the most in its head at once. Structural revision, a full-draft pass where the note on page 9 has to connect to the payoff on page 94, the kind of long-horizon work where you would rather wait a few extra seconds and get the deeper read.
None of this is about one tier being smart and another being dumb. It is about matching capability to stakes, the same way you would not send a table-read draft to a proofreader or a typo sweep to your most expensive script consultant.
The cost intuition
Plans with hosted passes carry a monthly allowance, and each pass draws against it in proportion to the model behind it. The fast tier is the baseline at 1×. Next to every tier, the picker shows a draws-N× figure computed from the provider's published per-token rates: on the Claude ladder the balanced tier draws about 3× and frontier about 5×; on the GPT ladder, about 2.5× and 5×. The figure is derived from the same rate table that meters the pass itself, so the hint you see and the usage you draw cannot drift apart.
The practical version: running everything on frontier does not break anything, it just spends your allowance roughly five times faster than running everything on fast. Most writers land in the middle and step up deliberately.
What Free and Pro unlock
The fast tier is on every plan, including Free. Balanced and frontier unlock on Pro. Defaults follow the same shape: a Pro account that never touches the picker runs balanced, and a Free account runs fast.
The clamp works in the honest direction too. If you picked frontier on Pro and your subscription lapses, your passes fall back to your plan's default rather than erroring out or quietly serving a tier your plan no longer includes. And in the picker, tiers above your plan are greyed out rather than hidden, so you can always see what an upgrade buys before you make it.
The picker names names
One more deliberate choice: the picker shows the exact model each tier currently resolves to. Not "fast-v2," not a marketing gloss — the concrete model name, per provider, as the fixture above shows. Your pick covers every AI surface except typing suggestions: skill passes, inline edits, and chat all resolve the same selection identically, so the model shown selected is exactly the model that runs.
Model lineups change; providers retire models and ship new ones. When that happens, the name in the picker changes with it, because the picker reads the same model registry the product routes through rather than hand-kept copy. You should never have to wonder which model rewrote your scene. It says so, right where you chose it.