Pricing
One line per surface, in dirhams, published.
Seats for the people doing the work, usage for what the machines do, reserved capacity when it has to be yours, a contract when it has to be in your building. No quote form standing between you and a number.
AED 89/ seat / month
For the people doing the work.
- Spaces, standing instructions, connected systems
- Artefacts with provenance, versions and freshness
- Delegated, scheduled and batch work
- Audit trail and audit-pack export
- Arabic and English interface
Around 20% less billed annually. Model usage inside the Workbench is included up to a fair-use ceiling.
AED 5.51/ 1M input tokens
falcon-h1-arabic-34b. Output AED 16.53. Smaller models from AED 0.29.
For what the machines do.
- Priced per model, input and output apart
- Priority at 2×, async and batch at half
- Cached input at a tenth of the input rate
- Agent runs meter on the same wallet
- Prepaid AED, no committed spend
Free tier is 10 requests a month, no card. Long context meters on total tokens, the context window included.
AED 220K– 1.29M / year
When the capacity has to be yours.
- Reserved GPUs, predictable latency, no noisy neighbours
- Your own tuned adapter, served only to you
- Air-gapped installation inside your perimeter
- SSO, SCIM, custom retention, zero retention
- SLA with named support in Arabic and English
About $60K to $350K a year, set by the hardware reserved for you. A steady workload is cheaper on a contract than per token, and we will tell you which side of the line you are on rather than letting you find out.
| Model | AED in | AED out | USD in | USD out |
|---|---|---|---|---|
| falcon-h1-arabic-34b | 5.51 | 16.53 | $1.50 | $4.50 |
| falcon-h1-arabic-7b | 1.29 | 3.67 | $0.35 | $1.00 |
| falcon-h1r-7b | 1.29 | 5.14 | $0.35 | $1.40 |
| falcon-e-3b | 0.29 | 0.92 | $0.08 | $0.25 |
| khalij-embed-ar | 0.18 | — | $0.05 | — |
| Tier | Multiplier | Applies to | Reach for it when |
|---|---|---|---|
| Standard | 1× | Input and output | The default. Everything that is not latency-critical or deferrable. |
| Priority | 2× | Input and output | A person is waiting. Same models, a shorter queue. |
| Async and batch | 0.5× | Input and output | Work that can wait. Half price for the same output. |
| Cached input | 0.1× | Input only | A repeated prefix. A long system prompt, or a document you ask about more than once. |
Billing and payment
Priced in dirhams, and invoiced the way a UAE entity needs.
The commercial mechanics are usually where a sovereign purchase stalls, not the technology. A government entity cannot pay by card against a foreign invoice in dollars, and being told to try is how a procurement cycle dies quietly.
Everything is AED by default. Priced, metered and invoiced in dirhams, from a UAE entity, with a TRN on the invoice and the 5% VAT stated as a line rather than absorbed into a rate. USD invoicing is available on contract.
Self-serve runs on a prepaid balance. Top up by card or bank transfer and see the balance and burn rate before it matters. A balance that runs out refuses the next request with a clear error rather than continuing and appearing on a bill.
Contracts run on invoice and purchase order. Annual or quarterly, net terms, your PO reference on every invoice, and a named contact who can answer a finance query without opening a ticket.
AED, from a UAE entity
No FX exposure between your budget and the invoice.
TRN and VAT on the invoice
Stated as a line item, so the invoice is reclaimable without a follow-up request.
Purchase orders and net terms
Terms that match how public and regulated entities actually pay.
Caps, alerts, no surprise renewals
Spend caps per run, per agent and per organisation, and renewal notice before a charge.
Questions
What people actually ask.
Do I need seats to use the API?
No. The API and Agents stand alone on a prepaid wallet. Seats are for people using the Workbench.
What happens when the wallet runs out?
Requests are refused with a clear error rather than silently degraded, and nothing is billed after the fact. You can set a low-balance alert.
Why is output more expensive than input?
Because generating a token costs more to serve than reading one. Charging a single blended rate makes everyone reading long documents subsidise everyone generating long answers, and the long Arabic document is the work we are built for.
Why is long context priced separately?
Because the memory behind a 256K request is the real cost of serving it, and burying that in a flat token price means everyone with short prompts subsidises everyone with long ones.
Is the Workbench seat price per named user?
Yes, per named user per month. External collaborators invited as read-only, watermarked and expiring participants do not consume a seat.
Do you charge for the audit trail or the export?
No. Evidence is not an upsell.
What does a dedicated deployment cost?
It depends on the hardware reserved for you, so it is a contract rather than a list price. We will give you a number in the first conversation, not the fifth.
Can we pay in USD, or against a purchase order?
AED is the default and the reason many customers are here. USD invoicing and PO billing are both available on contract, with your PO reference carried onto every invoice.
Is VAT included in the published prices?
No. Prices exclude VAT, and UAE VAT appears as its own line on the invoice with our TRN, so it is reclaimable without asking us for a reissue.
Will anything auto-renew or auto-charge?
Not without notice. Annual seat subscriptions send a renewal notice before the charge, and the prepaid balance is only topped up when you top it up.