Live platform · Malaysia

Sovereign AI, hosted on home ground.

Six self-hosted models today, with more open-source models on the way — three ways to run them, served from data centers on Malaysian soil, under Malaysian jurisdiction. No data leaves the country unless you ask it to.

Sign up Sign in
How a request is served Your applications connect by API, private link or an on-premise GB10 unit to an SNS facility inside Malaysian jurisdiction, where model weights, inference, token metering, prompts and logs stay. Cross-border routing sits outside that boundary and is off by default. MALAYSIAN JURISDICTION YOUR APPLICATIONS API PRIVATE LINK ON-PREM GB10 <10 ms SNS FACILITY Model weights Inference Token metering Prompts & logs CROSS-BORDER ROUTING off by default How a request is served Your applications connect by API, private link or an on-premise GB10 unit to an SNS facility inside Malaysian jurisdiction, where model weights, inference, token metering, prompts and logs stay. Cross-border routing sits outside that boundary and is off by default. MALAYSIAN JURISDICTION YOUR APPLICATIONS API PRIVATE LINK ON-PREM GB10 <10 ms SNS FACILITY Model weights Inference Token metering Prompts & logs CROSS-BORDER ROUTING off by default
100%
Data hosted in Malaysia by default
<10 ms
Avg. network round trip, KL Valley
N+1
Power redundancy, grid + backup generation
6
Self-hosted models, more on the way
The roster

Six models, run on our own hardware.

We run six models today, tuned and monitored on our own infrastructure, so each one performs the way its benchmark promised — not the way a shared, oversubscribed endpoint delivers it. More open-source and self-hosted models are on the way.

Model
Parameters · Context
Pricing per 1M tokens
Kimi 2.7-Code
1T–A32B
262K context
RM2.80 in · RM13.90 out
StrengthsLarge MoE capacity with efficient active compute; reasons well across large, multi-file codebases
Best suited forAgentic coding, multi-file refactors, repo-wide reasoning
Step 3.7-Flash
198B–A11B
262K context
RM0.66 in · RM3.80 out
StrengthsTuned for low-latency responses without giving up MoE-scale capacity
Best suited forReal-time chat, latency-sensitive customer-facing apps
MiniMax M2.5
230B–A10B
205K context
RM1.10 in · RM3.90 out
StrengthsBalanced reasoning and generation at a cost-efficient active-parameter footprint
Best suited forGeneral-purpose assistants, retrieval-augmented generation
DeepSeek V4-Flash 0731
Flash
1.3M context
RM0.16 in · RM0.41 out
StrengthsDistilled for speed and lower cost per call, on a current dated release
Best suited forHigh-volume lightweight tasks — summarisation, classification, extraction
Qwen 3.8 27B
27B
1M context
RM0.62 in · RM7.70 out
StrengthsStrong multilingual grounding across Bahasa, Mandarin, Tamil and English
Best suited forMultilingual support and localisation workloads
Gemma 4 31B
31B
262K context
RM0.37 in · RM1.40 out
StrengthsCompact and efficient, built for low-latency inference on modest hardware
Best suited forEdge and on-premise GB10 deployments, lightweight classification

All six models are shown above. More open-source and self-hosted models are on the way. Tell us what you need →

How you run it

Three ways onto the platform.

Access is metered by need, not by a one-size plan. Most teams start pay-as-you-go, move to enterprise once usage stabilises, and add on-premise hardware once inference becomes part of a physical workflow.

Pay-as-you-go

Metered access

For individuals and small teams testing use cases before committing to volume.

  • Priced per token, per model
  • All six models, no minimum spend
  • Self-serve API keys, live usage dashboard
  • Top up by card or local bank transfer
Start using tokens
Hybrid

On-premise + platform

A GB10-based unit installed on your premises runs your agentic applications locally, while model inference is served from our Malaysian data centers over a private link. Your orchestration and sensitive context never leave the building; only model calls do.

  • GB10 edge unit, installed and maintained by us
  • Private link to nearest SNS facility
  • Agent state and application logic stay on-site
  • Usage billed monthly against contracted volume
Talk to us about hardware
Enterprise

Flat-rate usage

Flat-rate access to the full roster for organisations with steady, high-volume workloads.

  • High-volume calls under a fair-use ceiling
  • Dedicated throughput, priority queueing
  • Uptime SLA and named support contact
  • Annual contract, invoiced locally
Discuss enterprise terms
Where it lives

Built for data that isn't allowed to leave.

Home
ground

SNS facility, Malaysia

Primary facility. Model weights, inference, and token metering all run here — inside Malaysian borders, under Malaysian law.

Data
residency

Nothing crosses a border by default

Prompts, completions, and logs stay on local infrastructure unless a customer explicitly configures cross-border routing.

On-prem
option

Keep sensitive workloads off the wire entirely

The Hybrid plan puts an on-premise GB10 unit inside your own building, so application logic and context never have to leave it.

Local
ownership

Malaysian-operated, not a regional proxy

SNS is not a reseller of foreign endpoints. Hardware, staff, and support are based in Malaysia.

SNS FACILITY — MALAYSIA
SNS Network

Status: Operational

ComputeNVIDIA GPU clusters
NetworkRedundant fibre, dual carriers
PowerN+1, grid + backup generation
JurisdictionMalaysia
AccessAPI · private link · on-prem GB10
Now live

Bring inference home. Keep the data where it belongs.

  • Hosted in an SNS facility, under Malaysian jurisdiction
  • Six self-hosted models, tuned on our own hardware
  • Hardware, staff, and support based in Malaysia
Compare plans
SNS Network MyToken

Create your account

Sign up to start using MyToken — the platform is live, no waitlist.

By registering, you agree to our Terms of Service and Privacy Policy.