Compare

Owned, rented, local.

Who owns the model, and does it keep learning? One table, three answers.

Rented AI

Where it lives
In someone else’s building
Whose model
The vendor’s
Your data
Sent to the provider
Learns your work
No. Where it learns from use, it trains the vendor’s model
After go‑live
Changes on the provider’s schedule
The bill as use grows
Per seat or per token, rising with use
Who runs it
The provider
Local fit
One global default

Local AI

Where it lives
On your own machine
Whose model
A stock open model, the same for everyone
Your data
On the device
Learns your work
No. It remembers context
After go‑live
Stays as shipped
The bill as use grows
Flat, capped by the machine
Who runs it
You
Local fit
One general model

Owned AI, Eigna

Where it lives
Where you decide: your machines or your region
Whose model
Yours, with a signed Model Deed
Your data
On your machines with Local, in the region you choose with Hybrid and Cloud
Learns your work
Yes, from approved work only
After go‑live
Keeps learning, released only when better
The bill as use grows
Setup, then a fixed running cost. No token bill
Who runs it
We deliver, set it up and run it
Local fit
Your language, laws and terms
ComparedRented AILocal AIOwned AI, Eigna
Where it livesIn someone else’s buildingOn your own machineWhere you decide: your machines or your region
Whose modelThe vendor’sA stock open model, the same for everyoneYours, with a signed Model Deed
Your dataSent to the providerOn the deviceOn your machines with Local, in the region you choose with Hybrid and Cloud
Learns your workNo. Where it learns from use, it trains the vendor’s modelNo. It remembers contextYes, from approved work only
After go‑liveChanges on the provider’s scheduleStays as shippedKeeps learning, released only when better
The bill as use growsPer seat or per token, rising with useFlat, capped by the machineSetup, then a fixed running cost. No token bill
Who runs itThe providerYouWe deliver, set it up and run it
Local fitOne global defaultOne general modelYour language, laws and terms
Our reading of public material, checked 24 October 2026.

Where renting is still right

Cost as use grows.

Rented AI bills every token, so heavy users pay more the more they succeed. An owned model runs at a fixed cost until the machine is full.

  • For light use of the cheapest rented models, renting stays cheaper. We make no cost claim against it.
  • For email, office work and open questions, the best general model is rented to everyone. Keep it.
  • The case for owning is the work that defines you: quality on your task, control of every version and a cost you can plan.
UseMonthly cost
  • Rented: billed by use
  • Owned: a fixed fee per site or GPU

Illustrative shape, no figures on the axes.

Eigna and the alternatives

Every part is for sale. The combination is not.

Open weights, tuning platforms, machines on site and boxes at home are all for sale today. A model you own, that keeps learning your approved work and comes delivered and run, is not.

A dark monolith of two plates on a salt flat at dusk, the Figure lit through it
OwnedThe weights, the data and the experts belong to you.Also offered by Mistral Forge, Prem, Arcee, Fireworks and Ghost.
Keeps learningRetrained on approved work. Live only when it beats the current version.Sold as a tool for ML teams, or for the vendor’s own model.
Delivered and runLocal or cloud, on your hardware or ours. No ML team to hire.Mistral does it, for the largest firms.
EignaOwned, keeps learning, delivered and run. Together, only here.

Rented assistants

For example
ChatGPT, Claude, Gemini, Copilot
Whose weights
The provider’s
Learns your work
No, context only
Gate and rollback
On the provider’s schedule
Where it runs
Their cloud
Cost as use grows
Rises with use
General work
Best available

Tuning platforms

For example
Fireworks, Together, Cohere, Mistral
Whose weights
Yours, if tuned on open weights
Learns your work
Once per job, run by your ML team
Gate and rollback
You build it
Where it runs
Their cloud, some on site
Cost as use grows
Rises with use
General work
Same as the base

Local AI

For example
Ollama, AI PCs, Ghost
Whose weights
A stock open model
Learns your work
No, memory only
Gate and rollback
None
Where it runs
On the device
Cost as use grows
Flat, capped by the box
General work
Below the frontier

Owned, Eigna

For example
Eigna
Whose weights
Yours, with a Model Deed
Learns your work
From approved work, gated
Gate and rollback
Release gate, record, rollback in 60 seconds (designed)
Where it runs
Local, Hybrid or Cloud
Cost as use grows
Flat per site or GPU
General work
Below the frontier: keep a rented model for it
ComparedRented assistantsTuning platformsLocal AIOwned, Eigna
For exampleChatGPT, Claude, Gemini, CopilotFireworks, Together, Cohere, MistralOllama, AI PCs, GhostEigna
Whose weightsThe provider’sYours, if tuned on open weightsA stock open modelYours, with a Model Deed
Learns your workNo, context onlyOnce per job, run by your ML teamNo, memory onlyFrom approved work, gated
Gate and rollbackOn the provider’s scheduleYou build itNoneRelease gate, record, rollback in 60 seconds (designed)
Where it runsTheir cloudTheir cloud, some on siteOn the deviceLocal, Hybrid or Cloud
Cost as use growsRises with useRises with useFlat, capped by the boxFlat per site or GPU
General workBest availableSame as the baseBelow the frontierBelow the frontier: keep a rented model for it
  • Choose a rented assistantFor broad office work, open questions and light use. Minutes to start, the best general model, and the lowest cost when use is light.
  • Choose a tuning platformWhen you have an ML team that wants to run each training job, its tests and its releases itself.
  • Choose local AIFor a general model on one machine, at home or on a desk. It remembers context; it does not learn your work.

Our reading of public material, checked 24 October 2026. Vendors change their offers often; tell us if a cell is out of date.

Questions and answers

Is owned AI cheaper than rented AI?

For heavy, steady use of the mid and top tiers, a fixed running cost wins. For light use of the cheapest rented models, renting stays cheaper, and we make no cost claim against it. The stronger point is a cost you can plan.

Tokens keep getting cheaper. Does that change the case?

Unit prices fall, and bills keep rising as use grows: the price per token fell to 0.42 times while the bill more than doubled (Bain, September 2026). For the rest, the case is quality on your task and control of every version.

Open models will make this a commodity, won’t they?

We hope so. Better open models lower our cost, and we re‑base on the best one, through the gate. What does not become a commodity is your data, your sealed test set, your release record and our Formula for your field.

Start

See which column your work belongs in.

The fit check takes two minutes and recommends a run mode. Or bring your questions to a founder.