A sovereign gateway. On someone else's cloud.
Northern Inference is a Canadian router in front of AWS Bedrock and Azure OpenAI, with real custody records and a fail-closed Canadian tier. But its frontier models route outside Canada, and its Canadian tier tops out at Llama 3 and GPT-4.1-mini. Runars serves GLM 5.3, Kimi K3 and DeepSeek V4 from its own infrastructure in Montreal and Calgary.
Credit where due
Northern Inference is one of the more interesting Canadian entries: per-request chain-of-custody records, a Canadian tier that fails closed, BYOK custody that keeps provider billing in your own account, and dated evidence on logging and retention. If you want to route existing AWS or Azure spend through a Canadian control plane, that's a real product. If you want frontier models that never leave Canada, read on.
Data residency
Zero border crossings.
Northern Inference's Tier 3 keeps requests on AWS Bedrock ca-central-1 and Azure Canada East — real residency, on US-owned clouds. But check the menu before you commit: its Canadian routes cover Llama 3, Mistral, Mixtral, GPT-4.1-mini and embeddings. Every frontier model on its price list — GLM 5, Kimi K3, Claude, GPT-6 — is Tier 4, which runs outside Canada. Runars serves GLM 5.3, Kimi K3 and DeepSeek V4 from Montreal and Calgary, on every plan, from the first token.
- 1your app🇨🇦 Canada
- 2runars.ca/v1🇨🇦 Canada
- 3inference🇨🇦 Montreal · Calgary
- 4logs + prompt cache🇨🇦 Canada
border crossings: 0
Pricing
The rate you see is the rate you pay.
Northern Inference passes through upstream rates, converted to CAD at metering time with a volatility buffer of undisclosed size, refreshed up to 24 hours late, plus a 9% service fee on every credit purchase. Runars sets its own CAD prices per token — the number on this page is the number on your invoice.
Compatibility
Keep your code. Change one URL.
Northern Inference speaks the OpenAI and Anthropic APIs and publishes Claude Code setup docs — genuinely convenient, and so do we. Compatibility isn't the differentiator here; what's behind the base URL is. Theirs forwards to AWS and Azure. Ours is the inference.
client = Anthropic(
base_url="https://runars.ca",
api_key=os.environ["RUNARS_API_KEY"],
)
# OpenAI SDK works too — same key.Head to head.
The short version, row by row. Every Northern Inference claim links to a source at the bottom of the page.
Same models. In CAD.
Per 1M tokens, input / output. Runars prices are live. Northern Inference prices were listed in USD on 2026-10-21 and are converted at 1.4054 USD→CAD for comparison.
| Model | Runars (CAD) | Northern Inference (≈ CAD) |
|---|---|---|
| GLM 5.3 | $2.23 / $7.01 | Not listed |
| Kimi K3 | $5.65 / $28.23 | $4.22 / $21.08(Kimi K3 — Tier 4 (mixed regions)) |
| Deepseek V4 Flash | $0.26 / $1.04 | $0.27 / $0.72(DeepSeek-V4-Flash — Tier 4 (mixed regions)) |
| Deepseek V4 Pro | $1.62 / $4.87 | $2.45 / $4.89(DeepSeek-V4-Pro — Tier 4 (mixed regions)) |
| GLM 5 | $1.59 / $5.10 | $1.41 / $4.50(GLM 5 — Tier 4 (US)) |
Switch in one line.
Point your existing OpenAI or Anthropic SDK at Runars and keep your code. If we're not the right fit next to Northern Inference, we'll tell you.
base_url = "https://…"
base_url = "https://runars.ca/v1"Two minutes. No sales call.
Sources · checked 2026-10-21
- Northern Inference — pricing
- Northern Inference — models
- Northern Inference — platform
- Northern Inference — terms of service (§3 billing, §4 residency)
Northern Inference is a trademark of its owner and is referenced here for comparison only. Competitor details change; check Northern Inference's site for current terms. Spotted something out of date? Tell us and we'll fix it.