Managed LLM inference on open models, at a fixed monthly price
An OpenAI compatible inference API for production teams. Prepay a month of counted tokens on one open model, with your requests at once held for you and an uptime SLA in the contract. The price stays the same for the whole term, whatever GPU prices do.
- Price for the term
- Fixed
- GPUs, never spot
- Reserved
- Data retention
- 0 days
import os
from openai import OpenAI
client = OpenAI(
base_url=os.environ["SPHERON_BASE_URL"],
api_key=os.environ["SPHERON_API_KEY"],
)
client.chat.completions.create(
model="glm-5.3",
messages=messages,
)Change the base URL and key. Keep the rest of your code.
A monthly allowance, paid up front
Every plan is counted tokens a month on one model, with requests at once held for you and a price fixed for the term. Run GLM-5.3, GLM-5.3-Flash, DeepSeek V4.1 Flash or another open model.
Starter
Growth
Dedicated
Your price depends on the model, your monthly volume and the term. Send the form and an engineer replies within one business day with a quote in writing. Go past your allowance and requests keep working, billed at your plan's own rate.
How tokens count
Output costs the most to serve and cached input the least, so each kind counts by its weight. One rule then covers chat and long agent prompts fairly. On DeepSeek V4.1 Flash, cached input counts 0.03.
1,500 counted
7,000 counted
1,400 counted
- Starter holds a month
- 505K of these requests
- Growth holds a month
- 2.02M of these requests
Inference for production workloads
Shared plans for steady product traffic, dedicated inference endpoints for teams that need whole nodes. Either way you call one API and pay one monthly price.
Agents and coding tools
RAG and enterprise search
Customer support and chat
Batch and data pipelines
Your own fine tuned model
A token factory for your product
What every plan includes
One price for the term
Reserved GPUs, never spot
OpenAI compatible
Zero data retention
Full weights, labelled
A usage page
From request to a live plan
Request a plan
Try it on a trial key
Sign and prepay
Live within 2 business days
Tell us about
your workload.
Tell us what you run and roughly how much. An engineer replies within one business day with a plan, a quote and a trial key if you want to test first.
A reply within one business day
An engineer sizes the plan to your traffic and sends a quote.
Try before you commit
A capped trial key runs your real traffic first.
Nothing charged until you sign
One order form for the term, then the first invoice.
Prefer a conversation first?
Book a call with an engineerRequest a plan
* Required
Questions teams ask
Allowances, overage, models, uptime and data.
We run the open model, the GPUs and the serving stack, and you call an API. You get an OpenAI compatible endpoint, a monthly token allowance and an uptime SLA, with no GPUs, drivers or autoscaling to look after.
Yes. A Dedicated plan runs your model on single tenant nodes sized to your traffic, with the region pinned to the EU or the US and p99 latency in the contract. Starter and Growth share reserved GPUs across plan teams.
GLM-5.3, GLM-5.3-Flash and DeepSeek V4.1 Flash on every plan. Dedicated plans can also run another open model, or your own fine tuned weights.
Yes. Point any OpenAI SDK, agent framework or coding tool at the base URL with your key. Chat completions, tool calling, structured output and streaming work as you expect.
Per token pricing moves with the market and gives no capacity guarantee. A plan fixes your price for the term and holds your requests at once, so a busy hour on someone else's traffic never slows yours.
It depends on the model, your monthly volume and the term. Send the form and an engineer replies within one business day with a quote in writing. Prepaying a quarter or more takes 5% off.
Nothing stops. Counted tokens past the allowance are billed at your plan's own rate on the next invoice, and we suggest a bigger plan if it keeps happening.
Unused tokens expire at the end of each plan month. You can move down a plan at any renewal.
Yes, at each renewal. Open models improve every few weeks, and your plan can follow them.
Your plan's model answering requests. If a month falls below the promise in your plan, service credits come off the next invoice without you having to ask.
Prompts and answers are never stored or used for training. Dedicated plans can pin the region to the EU or the US. Ask us for the security pack and the DPA.
Prepaying lets us reserve GPUs for your plan at a known cost before your month starts. That is why your price never moves during the term.
Lock your inference price for the term
A month of counted tokens on one open model, paid up front, at a price that does not move.
