ClusterBid
Inference brokerage

Reserve production
inference throughput.

Dedicated endpoints for open-source and custom models. SLA-backed p99 latency, zero rate limits, routed on verified GPUs instead of shared pools.

See supported models↓
Why brokered inference

Token endpoints, reserved
on real GPUs.

Most token APIs sit on top of shared pools where another customer's spike is your latency spike. We reserve dedicated capacity on verified hardware and wrap it in a clean, OpenAI-compatible endpoint.

01

Dedicated hardware, not shared pools.

Your tokens land on GPUs that are yours for the contract. No noisy neighbours, no throttle under load, no variable latency.

02

Wholesale unit economics.

Providers price inference at the GPU-hour level; we broker the raw capacity and translate it into token-level contracts that make sense for your workload.

03

OpenAI-compatible, day one.

Drop-in endpoint. Swap your base URL. Keep your code. We terminate TLS in the same region as your users or your training cluster.

04

SLA-backed p99 latency.

Every quote carries an explicit p99 target and throughput floor, not a 'best effort' disclaimer. Breach triggers credits automatically.

Supported models

Models available on request.

Model
Hardware
Llama 3.3 70B
H100
Llama 3.3 8B
L40S
Mixtral 8×22B
H200
DeepSeek-V3
H200
Qwen 2.5 72B
H100
Custom / fine-tuned
your choice

Custom and fine-tuned models welcome. Every engagement starts as a quote, sized to your workload, region and latency targets. Spot and serverless variants available on request.