Where they differ

  • Where the data goes: with a cloud model, every prompt and every document is sent to the provider's servers, often outside the country. With Orchestra, the prompts, indexes and logs stay on your hardware. The data rules, country by country.
  • Cost: the cloud bills per token, so the bill grows with every conversation. Orchestra has a fixed annual licence that does not move with use, plus hardware you buy once. How Orchestra pricing works.
  • Who controls the model: a cloud provider updates or retires the model on its own schedule, and the service can be cut off by a political decision or a change in its policy. Your version of Orchestra is yours, and you decide when it changes.
  • Arabic: the large cloud models are strong in formal Arabic. We built Orchestra from scratch for Arabic, for formal Arabic and for the dialects as customers write them. We are measuring its dialect quality now and will publish the results.
  • General capability: the largest cloud models are far bigger than Orchestra and stronger on open-ended general tasks. Orchestra is built for enterprise work in Arabic: answering from your documents with sources, and running agents. Its scores on standard benchmarks are on the results page.
  • Getting started: the cloud starts in minutes with a credit card. Orchestra starts with a pilot on your documents that takes weeks, and needs one GPU node on your side.

When the cloud is the better choice

  • You are testing an idea and want a result this week.
  • Your use is low or occasional, so paying per use is cheaper.
  • Your data is public, or nothing restricts sending it abroad.
  • You need the largest general models for open-ended tasks unrelated to your documents.

And when Orchestra fits better

  • Your data is sensitive or regulated: banking, health or government.
  • Use is high and continuous, as in customer service and contact centres.
  • Your customers write in their dialects.
  • You cannot accept your work stopping because of a decision by a provider abroad.

A middle path

If you are not ready to run it on your own servers yet, we host Orchestra for you and bill per token, and you can move to the fixed on-premises licence later: the hosted option. To work it out with your own numbers: how to find the break-even point between cloud and on-premises.

Comparing with an open model you would run yourself? Read Orchestra or an open model.

Frequently asked questions

Is Orchestra as capable as the largest cloud models?

Not at everything. The largest cloud models are stronger on open-ended general tasks. Orchestra is built for enterprise work in Arabic on your own documents. The right judgement is a test on your tasks, not on general ones.

Can we start in the cloud and move later?

Yes. Many teams try the idea on a cloud model with non-sensitive data, then move on-premises once real data comes in or use grows. You can also start with our hosted option and move to the licence.

Does Orchestra need the internet to work?

No. On your premises, the model, retrieval, safety and logs all run on your hardware, and nothing needs a connection to an outside service.