Qloud/ TOKEN FACTORYOpen Models

OPEN MODEL ACCESS

One API. Every open model.

The stack that tunes private workloads runs the public fleet. Point an OpenAI-compatible client at Qloud and call any open model.

Browse the catalog
BASE URLOpenAI-compatible
https://api.qloud.sh/v1
client = OpenAI(
    base_url="https://api.qloud.sh/v1",
    api_key=QLOUD_API_KEY,
)

client.chat.completions.create(
    model="deepseek-v4-flash",
    messages=messages,
)

Migration is a base URL and a key.

Access / 01

Four ways to buy the same catalog.

Same models in every mode. What changes is who absorbs idle capacity and how tightly latency is bounded.

4 access modesSwipe to compare

01
Per million tokens

Serverless

No setup

Any model, one endpoint. Capacity is shared, so you pay only for tokens you consume.

  • No minimum
  • Shared fleet
  • Day-zero models
02
Per million tokens

Elastic

Private, still per-token

A private endpoint our agents tune to your traffic. Dedicated behaviour without paying for idle GPUs.

  • Private endpoint
  • Tuned per workload
  • Scales with demand
03
Per GPU-hour, by the minute

Dedicated

When latency is contractual

Reserved accelerators sized to your roadmap, scaling policy under your control.

  • Reserved accelerators
  • Negotiated SLA
  • Private and fine-tuned weights
04
Discounted per million

Batch

Deadline-tolerant

Offline work on spare capacity, for jobs where real-time latency is irrelevant.

  • Lowest rate
  • Millions per job
  • Evals, embeddings, backfills

Catalog / 02

Open weights, tuned infrastructure.

Models are added as they ship. Quantized variants are labelled, never swapped in silently for the checkpoint you asked for.

DeepSeek V4 Pro

ReasoningLong context

Frontier reasoning for hard, low-volume work

DeepSeek V4 Flash

Low latencyHigh volume

Distilled tier for interactive and batch traffic

Start / Open models

Bring the workload.
Keep the weights open.

Back to the foundry