Cited, not adopted
That number is measured by the people who run the model. It describes their service rather than ours, and it moves as the window rolls forward. Check it there, not here.
SLA
Availability
We do not own the compute, so we cannot sell you an uptime figure. What we can do is show you where the real number lives and commit to the parts that are genuinely ours.
The number you are looking for
The model service publishes its own availability. Over the window currently shown on its status page (July to October 2026), the DeepSeek V4.1 Flash API service is listed at 99.69%. The live figure is at status.deepseek.com.
The honest position
The availability of the model itself is the provider's to publish and yours to verify. We are not going to blur that line to look bigger than we are.
Why we will not restate it as ours
Why not just print a number
It would be easy to print an availability figure in our footer and let you read it as a promise. We are not going to, because the compute, the network and the model all sit outside this organization. If we published a number we do not control, the first serious outage would turn it into a false statement — and you would have made a procurement decision on it.
What we do commit to
The parts of the experience that are actually ours, written down so you can hold us to them.
We answer
An acknowledgment on any incident report within one business day, from a person, not an autoresponder.
We tell you when we do not know
If the cause is upstream and we have no estimate, we will say that rather than invent a timeline.
We tell you first
When we can see an upstream problem affecting you, we will reach out rather than wait for your ticket.
One named owner
Your account has a contact who owns the incident through to resolution, rather than a shared inbox that rotates.
An honest account afterward
What happened, what it affected, and what changed as a result.
Credits, and what they would not cover
Service credits
Some procurement processes require a credit regime in the contract — a defined remedy when availability drops below a threshold. We will negotiate one into your agreement and set the threshold against the provider's published figures, with the measurement window and the source of truth named in the document. We would rather do that than publish a generic formula on a web page that we cannot honor for every customer.
What is not covered
Anything on your side of the connection: your network, your application, your key handling, and any change you make to your integration. We will help you diagnose those, but they are not availability failures on our side, and a credit regime that pretended otherwise would be priced for it.
Before you raise a ticket
Check the status page first — if the provider is reporting an incident, we already know, and you will get a faster answer by asking us for the current state than by describing the symptom from scratch. Then tell us what you sent and what came back, and we will take it from there.
Last reviewed: 2 October 2026. The availability figure above is quoted from the model provider's public status page and is not a commitment by us. Contractual commitments are those in the agreement signed with your organization.
SLA · WHAT WE PUT IN WRITING
BEFORE YOU SIGN
We tell you what we can promise
and what we cannot
IN THE CONTRACT
Response and availability terms
agreed per contract, not a blanket figure
IF IT BREAKS
A named person, and a route
escalation in writing, not a queue
- Availability figures come from the model provider
- Their public status page is the reference
- No number we could not stand behind
Need an availability clause in writing?
Tell us the threshold and the window your procurement team needs, and we will draft it into the agreement.