Every plan starts with a 14-day trial: 1,500 compute units, no card.
One scientist, with a machine of their own.
3,000 compute units a month, then $30 per 1,000
A group sharing one pool of compute. Minimum 3 seats.
10,000 compute units per seat, pooled, then $30 per 1,000
Pharma and core facilities, on your own model endpoint.
No metered compute: your endpoint, your inference bill
Prices are per month and exclude tax. Compute units are pooled across an organisation on Lab and above.
FAQ
One number covering all the work Marigold does on your behalf: model calls, verification, literature search, parsing, sandbox time, storage and egress. Billing in tokens is incomprehensible, and billing per message hides a large variance in what a message really costs, so it is all converted into a single unit you can reason about.
Every account starts on a 14-day trial with 1,500 compute units and no card. It ends rather than renewing, and it comes with 1 project, 2 GB of storage and 3 deep research reports, with batches switched off.
Researcher and Lab have overage on by default and carry on at $30 per 1,000 units. You can also buy top-up blocks of 1,000, 5,000 or 20,000 units on top of your plan. There is a per-project spend cap and an organisation-wide kill switch if you would rather stop than spend.
Yes. When the tier changes, the new allowance applies to the new period straight away, so you are not waiting on a renewal date for what you have paid for.
Because the compute is pooled rather than per person. Each seat adds 10,000 units to one shared balance, so a group with uneven workloads is not paying for allowances that go unused. The minimum is 3 seats.
We work with pharmaceutical companies and research institutions worldwide. Let's discuss your needs.