Sovereign inference · us · eu · in

Sovereign frontier open-model inference.

Llama-3.1-405B, DeepSeek-V3, and the frontier open-source models — served in your region on Blackwell-class GPUs, with the contracts your Legal team asks for.

// three regions · live at launch · routing enforced in code
us · iadeu · fra · pinnedeu · mad · pinnedin · bom
live · routing enforced in code

We sign your DPA, your BAA, your sovereign-cloud agreement.

Region-resident routing enforced in code, not in a policy PDF.

Contract-able terms on our paper — counsel-reviewed, ready before our next call.

Subprocessor list published, with 30-day change notification.

Read our trust center →
// what we sell

Three SKUs, one OpenAI-compatible API.

Shared inference for builders. Single-tenant deployments for regulated buyers. Managed fine-tuning where the launch-set models need to learn your data. Same SDK, same region pin, zero code change to move between them.

sku 1

Inference API

OpenAI-compatible HTTP API. Five launch models. Per-token billing. $10 free on signup, no card required to start.

  • Pricing per 1M tokens
  • Regions us · eu · in
  • SDK OpenAI
  • Free credit $10
Get an API key
sku 2

Dedicated Deployments

Single-tenant on B200 / B300 nodes. Per-node-month pricing with 3- and 12-month commits. Same protocol as the API — base-URL change to migrate.

  • Pricing per node-month
  • Term 3 or 12 mo
  • SLA 99.9% / 99.99%
  • Isolation single-tenant
Talk to sales
sku 3

Managed Fine-Tuning

LoRA fine-tuning on Llama, Mixtral, and DeepSeek base models. JSONL upload, hyperparameter defaults with override, cost preview before you commit.

  • Pricing per training run
  • Method LoRA
  • Output adapter
  • Deploy one-click
Read the docs
// frontier card

405B-class served well.

Most inference clouds top out at 70B. We lead with the largest open models because Blackwell-class hardware is the moat.

llama-3.1-405b-instruct
Llama-3.1-405B
parameters405B dense
context128k
regionsus · eu · in
throughput≥ 25 tok/s
price · output$5.00 / 1M
deepseek-v3
DeepSeek-V3
parameters671B MoE · 37B active
context128k
regionsus · eu · in
throughput≥ 50 tok/s
price · output$1.10 / 1M
hermes-3-405b · beta
Hermes-3-405B
parameters405B dense
context128k
regionsus (beta)
throughput≥ 25 tok/s
price · output$5.00 / 1M

See all six launch models →

// why clusterbid

Three things we commit to.

Read the positioning doc for the long form. The short version is what we are, and what we deliberately are not.

Sovereign by default.

Region-resident inference in US, EU, and India is the baseline, not a premium tier. Data stays in the region you pick. Routing enforces this in code.

Read more →

Frontier open models, served well.

Llama-3.1-405B, DeepSeek-V3, Hermes-3-405B. The biggest open models at production latency on Blackwell. Curated for license, capability, and a documented buyer profile.

See the catalog →

Production-grade, not hobbyist.

SLAs in writing. Logs, metrics, traces from day one. Status page, incident comms, scheduled maintenance, change management. We sell tokens and uptime, not hours of SSH.

Read the SLA →

What our trust center contains.

DPA, BAA, sovereign-cloud agreement, subprocessor list, SOC 2 status, residency promise, security whitepaper, vulnerability disclosure.

Open the trust center →

PostureStatus
SOC 2 Type IReadiness · Q3
SOC 2 Type IIAudit · Q4
GDPR — DPAAvailable · on paper
HIPAA — BAAAvailable · on request
India DPDPAddendum available
EU AI ActWhitepaper published
get started

Start with $10 free credit, or sign a contract today.

Same SDK, same region pin — move between the API, dedicated deployments, and fine-tuning with zero code change.