Onde Inference / Pricing

On the device when you can. In the cloud when you need to.

Build with the SDK for free. Connect an app to add fleet tools. Call Onde Cloud when a task needs more than the device.

Device plans and cloud usage are billed separately, and both roll up to your organization.

See cloud rates

1. The SDK

Free to build with. Inference runs on the device. No account needed to start.

2. Connect an app

An app key adds Pulse, models assigned from the dashboard, and active-device counting. Priced per active device.

3. Onde Cloud

The same app key reaches one OpenAI-compatible endpoint. Pay per million tokens. You pick a tier, not a provider.

Start clean. Scale hard.

Plans count active devices reporting to an app. Pick the one that matches the stage you are in.

For the dreamers.

Garage

$0/mo

Up to 5 active devices

Start small. Ship fast. Keep full SDK access from day one.

Up to 5 active devices reporting to an app
Full SDK access
On-device inference included
Pulse usage and latency for your devices
Community support

Build for free

For the makers.

Pro

Most chosen

$49/mo

+ $0.05 per active device

When the prototype becomes a product, Pro keeps rollout simple and operations sharp. Device plans are billed separately from cloud usage.

Everything in Garage
Billed per active device
Models assigned and updated from the dashboard
Pulse across your whole fleet

Go Pro

For the icons.

Scale

Custom

Talk to us

When performance is part of the brand, we tune the stack to the hardware and the business.

Everything in Pro
NPU-specific optimization
SLA
Private control plane

Contact sales

Pay per million tokens.

One OpenAI-compatible endpoint. Ask for a tier by id; Onde decides what answers. Personal organizations pay as they go.

USD per 1M tokens
Tier	Input	Cached input	Output
Onde Flux`onde-flux`	$0.17	$0.005	$0.33
Onde Aura`onde-aura`	$0.18	$0.035	$0.58
Onde Mini`onde-mini`	$0.12	$0.030	$0.46
Onde Flare`onde-flare`	$0.35	$0.035	$1.40
Onde Prism`onde-prism`	$0.35	$0.070	$1.40
Onde Air`onde-air`	$0.46	$0.12	$1.85
Onde Fast`onde-fast`	$1.15	$0.12	$5.75
Onde Nova`onde-nova`	$1.65	$0.30	$5.10
Onde Orbit`onde-orbit`	$1.65	$0.30	$5.10
Onde Pro`onde-pro`	$2.30	$0.58	$9.20
Onde Zenith`onde-zenith`	$2.30	$0.29	$6.90
Onde Apex`onde-apex`	$2.45	$0.23	$5.10
Onde Balanced`onde-balanced`	$3.45	$0.35	$17.50
Onde KKK`onde-kkk`	$3.45	$0.35	$17.50
Onde Large`onde-large`	$5.75	$0.58	$29.00
Onde Lumen`onde-lumen`	$6.00	$1.20	$30.00

For individuals and small teams.

Pay as you go

Usage is metered per request and totalled each month on your Billing page. Personal organizations start under a spending cap; add a card to raise it. Set a monthly budget with a hard limit, and a daily limit on each app.

Get an API key

For platforms and teams.

Business

The setup behind siGit Code and SplitFire AI: a team organization that gives every end user an app of their own.

A team organization with owners, admins, and members, and one bill
One app per end user, provisioned through the Onde Cloud SDK, so each user’s spend is capped and visible
Per-app daily limits and a monthly hard limit for the organization
Volume pricing, by agreement
Invoiced monthly instead of by card, with an optional committed spend

Talk to sales

Straight answers.

Why on-device?

Because speed and privacy are product features. On-device inference means no network round-trip and no server bottleneck, works offline, and scales without turning every session into infrastructure cost.

When should I use the cloud?

When the task is larger than the device: long contexts, heavy reasoning, or a device that is too old to run the model. Your app keeps running on-device by default and calls Onde Cloud with the same app key when it needs to.

How is cloud usage billed?

Per million tokens, at the rates above, split into input, cached input, and output. Every request is metered against your organization, and the month’s total shows on your Billing page. Device plans are billed separately.

Can I cap spend?

Yes. Set a monthly budget with a hard limit for the organization, and a daily limit on each app. A request that would pass a hard limit is refused.

Is it secure?

Yes. Inference runs on the device, so sensitive data does not need to leave the hardware. That reduces exposure by design. Cloud requests are encrypted in transit and authenticated per app. For teams that need more, Onde supports signed updates, controlled telemetry, and private infrastructure options at Scale.