Singulus.ai

Models

The Singulus-1 family.

Energy-efficient, privacy-first language models hosted entirely in Canada. Transparent per-million-token pricing. No data retention. No surprises.

Singulus-1 Nano

Lightweight & fast for edge deployment

3B32KEN · FR

Compact model engineered for on-device and edge inference. INT4-quantized weights ship under 2 GB and start generating in under 100 ms.

61.2

MMLU

38.5

HumanEval

42.1

GSM8K

  • Sub-100ms first token
  • INT4 / INT8 weights
  • Runs on consumer GPUs
  • Offline-capable

Input

$0.10 / 1M

Output

$0.30 / 1M

Singulus-1 Pro

Production-ready reasoning & coding

24B128KEN · FR · DE · ES

Balanced model for complex reasoning, document analysis, and software engineering. Our recommended default for most production workloads.

78.4

MMLU

72.1

HumanEval

76.8

GSM8K

  • 128K context window
  • Tool & function calling
  • Strong on HumanEval+
  • JSON-strict output mode

Input

$1.20 / 1M

Output

$3.60 / 1M

Singulus-1 Sovereign

Enterprise Only

Maximum capacity, sovereign infrastructure

120B (MoE)200KEN · FR · 12 more

Our largest model — a 120B mixture-of-experts for organizations that need state-of-the-art reasoning under full Canadian sovereignty.

85.1

MMLU

86.3

HumanEval

90.2

GSM8K

  • Mixture-of-Experts architecture
  • Dedicated sovereign capacity
  • VPC & on-prem deployment
  • Custom fine-tuning

Input

$4.50 / 1M

Output

$13.50 / 1M

Comparison

Compare models

Pick the right size for your workload.

ModelParamsContextMMLUHumanEvalGSM8KInput / 1MOutput / 1M
Singulus-1 Nano3B32K61.238.542.1$0.10$0.30
Singulus-1 Pro24B128K78.472.176.8$1.20$3.60
Singulus-1 Sovereign120B (MoE)200K85.186.390.2$4.50$13.50

Pricing built on trust, not extraction.

Pay only for the tokens you use. No data monetization. No hidden retention. Enterprise volume pricing and dedicated capacity available for teams running production workloads in Canada.

  • Per-million-token billing, billed monthly
  • Volume discounts above 50M tokens
  • Dedicated VPC & on-prem options
  • No training on your inputs — ever