Open-weight models
Language, image and video models from the major open families. ComfyUI workflows can be packaged as an API service.
For example · DeepSeek · Qwen · GLM · Kimi · Llama · Wan · ComfyUI workflows
Your models on dedicated GPUs. We run the stack.
Private LLM deployment, end to end: we size the hardware, source it or deploy on servers you already have, put open-weight or your own models into production, and operate them under a 99.9% monthly availability SLA.
- Open-weight or your own weights
- Hardware sourced by us or by you
- Operated by RouteAPI after delivery
- 99.9% monthly availability
- YouYour applicationCalls an OpenAI-compatible endpoint
- RouteAPIOperationsMonitoring, upgrades, incident response
- RouteAPIModel servingInference engine, quantization, scaling
- RouteAPI or youGPU serversSized to your model and traffic
Bring a model. Any open or private weights.
If the model licence permits it, we can put open-weight and your own models into production.
Your own weights
Fine-tuned checkpoints, LoRA adapters, or models your team built in-house. You provide the weights; we deploy and serve them.
You train · we deploy and operate
This service does not include model training or fine-tuning. If your project requires it, we will say so during the initial requirements call.
From sizing to on-call, one team
The same engineers handle sizing, deployment and operations after go-live, including incident response outside business hours.
Scope
Model, traffic profile, latency target, context length and data constraints, turned into a GPU count and a serving plan.
You get · Sizing sheet and quote
Hardware
We procure and host the servers, or we deploy onto servers or a cloud account you already own.
You get · A confirmed hardware plan
Deploy
Inference engine, quantization and scaling set up, then load-tested against your traffic profile before go-live.
You get · An OpenAI-compatible endpoint
Operate
Monitoring and alerting, engine upgrades, model version changes and incident response, covered by a 99.9% monthly availability SLA.
You get · A named team on call
Your servers or ours
Either way, we deploy and operate the software layer and carry the SLA.
| We source the hardware | You already have it | |
|---|---|---|
| Who owns the servers | Procured by RouteAPI for your deployment | You, on-premises or in your cloud account |
| Where it runs | A data center we arrange, in a region you choose | Your data center or your cloud region |
| What you pay for | One monthly fee for hardware and operations | A deployment fee plus an operations fee |
| Usually right when | You want to avoid an up-front hardware investment | You already have capacity, or data cannot leave your environment |
Your deployment stands alone. Public models are there if you want them.
The deployment belongs to you and runs on its own, without passing through the RouteAPI platform. RouteAPI also operates a gateway to 70+ public models, so the two can be connected when your workload calls for it, on terms agreed with you.
- By default, all traffic stays on your dedicated deployment
- Route harder tasks to GPT, Claude or Gemini, if you choose to
- You decide the failover plan to public models, or choose none at all
When this is the right call
We are a focused team. Below is where this service is a good fit, and where another option will serve you better.
Good fit
- Steady, high volume where dedicated GPUs cost less than per-token pricing
- Data or weights that must stay isolated on hardware dedicated to you
- A trained or fine-tuned model that now needs to run in production
- No in-house infrastructure team to keep GPUs running around the clock
Not a fit
- Training or fine-tuning. This service covers deployment only.
- Low or spiky traffic. Our pay-as-you-go API will cost less.
- Clusters of thousands of GPUs. A large GPU cloud provider is better placed.
Frequently asked questions
Send us your model and estimated traffic. We will come back with a sizing estimate and an initial quote.
Talk to our teamYes. We can deploy on servers you own, on-premises or in your cloud account, or source and host the hardware for you.
Monthly availability of your deployed endpoint. How downtime is measured and how service credits apply are set out in your deployment contract.
No. Your deployment runs on its own. If you also want access to public models or a failover plan, we agree that with you separately.
Only the RouteAPI engineers operating your deployment. We do not use your weights, prompts or outputs to train models.
Quoted per deployment after sizing, based on hardware, model and the level of operations you need. There is no per-token charge on your dedicated deployment.