CritiqueCode All In. Bring nothing. We run it.

$10 a month. A cloud machine for your coding sessions. $20 of included model inference, measured at provider cost. That is the entry plan for CritiqueCode All In, the bundle we’re announcing today.

For regular cloud use, $30 a month includes $60 of inference, a higher session allowance, and a priority queue on cloud machines. Both plans use Critique’s provider account. You bring the task.

The agent, the machine, the model

Starting a cloud coding session can mean choosing a provider, creating an account, finding a key, adding model credit, and checking how the machine itself is billed. These are useful controls when you want them. They are also several decisions between an idea and a working branch.

All In brings those pieces together. CritiqueCode is the author agent: it writes code, runs tools, and takes its work through independent review and verified repair. The cloud VM gives that session its own workspace. Included inference covers the model calls on Critique’s account.

You choose a monthly plan and work within its allowances. The VM is included separately; it does not eat into the dollar allowance reserved for inference. Model choice still matters, because different models use that allowance at different rates.

Two allowances, with a clear stop

A short fix and a long refactor can have very different costs. Counting prompts does not explain that difference. Counting sessions alone does not either: one session might be a small change; another might involve repeated model calls, tests, and repairs.

All In has two monthly caps: a number of cloud sessions and a dollar allowance for inference. Whichever cap you reach first ends that month’s bundled usage. If inference runs out while you have sessions left, those sessions do not come with more included inference. If sessions run out first, remaining inference does not create more cloud sessions.

All In is a monthly subscription you can cancel at any time. There are no automatic charges beyond the plan. We will publish the upgrade path and any optional inference top-up terms before launch. An inference top-up will not reset a session cap. The allowance resets with the next monthly billing period.

Pick a model for the work

The announced lineup brings efficient coding models and models for deeper reasoning into the same subscription. Choose for the task at hand, with the available routes visible when you start a session.

When capacity allows, we also plan to offer GPT-5.6 Sol, Claude Opus 5, Kimi-K3. These are additional options when available, rather than guaranteed monthly access. The base bundle should be a useful purchase on its own.

Local Code stays free

All In is a way to pay for cloud Code. Running CritiqueCode on your own machine still has no execution fee. Use a provider key you already have, or pay for model usage through Critique’s separate inference balance. Local BYOK use does not require a Critique account.

Local sessions never touch your All In balance. Standard CritiqueCode cloud is also a separate option: a VM per coding session, with BYOK or Critique inference billed separately. The paid VM rate will be shown before paid usage begins.

And CritiqueCLI stays the independent verifier

CritiqueCLI attaches to the coding agent you already use and independently checks its proposed change. It is a different product from the CritiqueCode author agent. Free locally. $0.15 per cloud stage; model usage separate. Starter $15 for 100 stages, Builder $75 for 500 stages, Team $300 for 2,000 stages. No seats.

CLI cloud stages keep execution and model costs separate on the receipt. All In bundles cloud execution and inference into a monthly offer. A stage pack is not an All In subscription, and neither one is a deposit into your separate inference balance.

Why start with twice the inference?

We want the first cloud sessions to be worth doing. Including twice the subscription price in inference at provider cost is a deliberate subsidy. We are funding the difference, along with the machine, to make All In an accessible way to try regular cloud coding.

This is a starting offer, not proof that the economics work. Real usage will tell us how long sessions run, which models people choose, and what happens when an allowance runs out. We will watch whether people upgrade, buy additional inference, or leave. We do not have those results yet.

That is also why the session count needs a compute estimate and a clear duration rule before launch. The monthly amount should be simple to understand, and the allowance behind it should be something we can deliver.