Pricing — Arclode

Pricing

Usage-based. Nothing wasted.

You pay for the generations you run, at cost-plus — and one credit balance works across Arclode and the rest of Zatabase.

Get started

Pay per job

Credits draw down by the realized cost of each generation. No subscriptions, no idle GPU bill.

Top up with Stripe

Add credit with a card through Stripe. Your balance is held by Zatabase — one wallet across every product.

Earn it back

Run a compute node and your earnings land in the same balance you spend from.

The billing unitLaunch estimates — final rates publish at GA

1 Arclode Unit (UNIT) = $0.01

1 UNIT ≙ ~20 IFL-R1 (RTX 5090-class) GPU-seconds — index v1, provisional, re-indexed monthly

Jobs are billed at realized provider cost plus a 15% network fee; the peg is re-indexed monthly.

Per-capability rates

Image generation

inference:image:generate

fal-ai/flux-pro/kontext/max/multiper image

10 units · $0.10

Typical job ≈ 10 units (≈$0.10)

fal-ai/flux-2-proper megapixel

4 units · $0.04

Typical job ≈ 4 units (≈$0.04)

fal-ai/nano-bananaper image

5 units · $0.05

Typical job ≈ 5 units (≈$0.05)

Image → 3D mesh

inference:3d:image-to-mesh

fal-ai/trellisper mesh

3 units · $0.03

Typical job ≈ 3 units (≈$0.03)

fal-ai/hunyuan3d/v2per mesh

19 units · $0.19

Typical job ≈ 19 units (≈$0.19)

Text → 3D mesh

inference:3d:text-to-mesh

fal-ai/hunyuan-3d/v3.1/pro/text-to-3dper mesh

44 units · $0.44

Typical job ≈ 44 units (≈$0.44)

Video generation

inference:video:generate

fal-ai/kling-video/v2.1/pro/image-to-videoper video second

12 units · $0.12

Typical job ≈ 60 units (≈$0.60)

fal-ai/bytedance/seedance/v1/pro/image-to-videoper video second

14 units · $0.14

Typical job ≈ 70 units (≈$0.70)

fal-ai/minimax/hailuo-02/standard/image-to-videoper video second

6 units · $0.06

Typical job ≈ 36 units (≈$0.36)

Texture generation, Auto-rig, Text generation, Music generation, Sound effects, Text-to-speech, and Transcription pricing publishes when those capabilities go live.

Trust tiers

Every job carries a tier, set by one field on the request. They differ by which machines see which part of your work and what you can verify afterwards. Open is the default and is priced at the rates above; Trusted costs more because a share of every job is reserved for vetted capacity — and less than Private because the rest of it isn't; Private costs the most because it needs confidential-computing hardware end to end.

Open · default

Anonymous providers, best-effort

Jobs route to anonymous contributor machines on any hardware. Your job runs whole on one of them, so the person running that machine can read your prompt and your output. This tier is explicitly not private, and nothing about it should be read as a confidentiality control.

Right for public or synthetic data, batch evaluation, benchmarking, and any prompt you'd be fine publishing. Wire value:best-effort.

Trusted

Vetted where it counts, cheap in the middle

A Trusted job splits into slices and the machines mix.The slice that reads your input and the slice that produces your output go to vetted operators — our own machines, bare metal we rent and run our own image on, RunPod, and operators we hold under contract — while the middle slices are filled with cheap anonymous capacity. That mix is why Trusted prices under Private; the default is a solid majority vetted and you can dial it to every slice.

A contributor on one middle slice sees intermediate math, not your prompt or your result, and extracting anything from it takes a deliberate attack — a small risk priced in, not a guarantee. Vetting is the bar, not hardware: a vetted operator on a consumer card qualifies. Right for real customer workloads. Wire value:trusted.

Private · premium

Encrypted from the host, and signed

Runs only inside a confidential VM on a confidential-computing GPU: system memory encrypted, GPU memory encrypted, and the traffic across the link between them encrypted under a session the hardware negotiates.The operator of that machine — us included — cannot read your prompt, your weights, or your output. These jobs never touch an open contributor machine.

The hardware also proves what it did: the launch is measured and the GPU signs an attestation, which must pass before any weight or prompt is shipped. So the output is attributable and tamper-evident and you verify what executed instead of trusting us. It can carry a whole payload, at a premium over the rates above. Wire value: confidential.

Both upper tiers are earnable on the provider side: Trusted by getting vetted — no special hardware, a consumer card counts — and Private by having confidential-computing hardware, which gates on the machine rather than on who owns it. Run a node and the jobs your machine qualifies for pay accordingly.