Last week we shipped Synthesis, Compare, and Debate: three ways to turn every model's answer into one you can trust. Today's release is about everything around the answer: your unused credits now roll over to the next month, you can write instructions once and apply them to every model, your best prompts are one click away, and you always see what a message costs before you send it. There's also a new plan.
Your credits roll over now
The change I'm happiest about is the least flashy one. Starting this month, whatever you don't use of your monthly credits carries into the next month, up to a full month's worth on top of your regular allotment.
There's nothing to enable and nothing to claim. If you had credits left over last month, they're already in your balance. You can enter a month with up to double your plan's credits. Add-on credit packs were already exempt from expiry; now your plan credits get the same respect.
The reasoning is simple: you paid for those credits. A quiet month shouldn't mean the money evaporates. Most AI subscriptions (including the single-model ones and most aggregators) reset you to zero every cycle and keep the difference. We'd rather you keep it.
Personas: write your instructions once
If you've ever pasted the same preamble into a chat for the tenth time ("you're advising a bootstrapped founder, be direct, skip the caveats"), personas are for you.
A persona is a reusable set of instructions: a role, a tone, standing context about you or your work. Pick one from the + button in the prompt bar: choose from 57 pre-built templates across marketing, engineering, writing, research, and more, or create your own in Settings → Personas. From that point, every model in that chat follows it. That's the part that makes personas on Deepest different from custom instructions on a single-model app: you write the instructions once and they steer GPT, Claude, Gemini, and whatever else you've selected, simultaneously, in the same conversation.
You can switch a chat's persona mid-conversation, and the change applies from your next message. Chats without a persona behave exactly as before.
Saved prompts, one bookmark away
Some prompts are too good to retype. Open the + button in the prompt bar and choose Saved prompts: "Save current" keeps what you've written, and next time you search and insert it with one click.
It's deliberately simple: no folders to manage, no templates to configure. The first line of your prompt becomes its title, and the whole library is searchable. Your best prompts compound the same way your credits now do.
Know the cost before you send
Every model on Deepest has a published credit price: budget models cost 1 credit per message, flagships like Claude Opus cost 30. Those prices now show right in the model picker, where you're actually choosing what to run.
And the composer watches your back without nagging: it stays quiet for everyday messages, but warns you before an unusually expensive send, when your balance runs low, or if you don't have enough credits (before you send, not after). Hover the warning for the full per-model breakdown. There's also a new wand button that rewrites your prompt for clarity before you send. You choose whether to use the suggestion or keep yours. It's free on every plan.
Between published per-model prices, the pre-send preview, and rollover, the goal is the same: you should never be surprised by your usage. Credit anxiety is the number-one complaint we hear about other AI platforms, and we want it to be structurally impossible here.
A new plan in the middle, and a rename
Until now there was a big jump in our pricing: Plus at $20/month with 5,000 credits, then straight to $99/month. A lot of you sat exactly in that gap: more than a casual user, not quite an agency.
So there's a new plan: Pro, $49/month, with 12,000 credits, unlimited models per prompt, and priority support. It's the plan for people who use Deepest every working day.
And the top plan grew into its role: Max is now $199/month with 65,000 credits (the best per-credit rate on Deepest, because the top tier should reward volume), plus bring-your-own-API-keys and priority support. It's the plan for people (and small teams sharing a login they shouldn't be sharing: get the Team plan) who run serious daily volume across many models.
Two smaller things worth knowing
A live status page. deepest.app/status shows the real-time status of the major AI providers, pulled from their official feeds every minute. It exists because of a truth we've built the whole product around: any single provider can go down, and when one does, the other 300+ models on Deepest keep answering.
Install Deepest as an app. Deepest is now installable. Look for the install icon in your browser's address bar (or "Add to Home Screen" on your phone). You get Deepest in its own window, one tap from anywhere.
Frequently Asked Questions
Do I need to do anything to get credit rollover?
No. It's automatic for all paid plans. If you had unused credits last month, they're already reflected in your balance this month. You can check it any time in Settings → Billing, or by hovering the cost counter in the prompt bar.
Is there a limit to how much rolls over?
Yes: one month's allotment. On Plus (5,000 credits/month) you can carry over at most 5,000, so the most you'd ever start a month with is 10,000. Rollover looks back one month and doesn't compound beyond the cap. Add-on credit packs are separate and never expire.
What happened to the $99 plan?
It became Max: $199/month with 65,000 credits, more than double the credits at a better per-credit rate ($3.06 vs $3.30 per 1,000), plus bring-your-own-API-keys and priority support. With rollover, a Max subscriber can enter a month with up to 130,000 credits available.
How are personas different from just pasting instructions into the chat?
Two ways. First, you write them once and reuse them across every chat instead of re-pasting. Second, and this is the Deepest-specific part, one persona steers all the models in a chat at the same time. Ask five models the same question and every one of them answers in your voice, with your context, under your rules.
