Serverless inference at scale.
OpenAI and Anthropic SDK compatible. Deploy chat and coding model inference instantly. Pay per 1M tokens, scale automatically. No infrastructure to manage.
Live model catalog.
Access our full roster of chat and coding models. Real pricing updated as rates change. OpenAI and Anthropic SDK compatible.
Base URL: https://runars.ca/v1
| Model | Input ($/M) | Output ($/M) | Context |
|---|---|---|---|
| Deepseek V4 Flash🇨🇦 | $0.2593/M | $1.0371/M | 1M |
| Deepseek V4 Pro🇨🇦 | $1.4328/M | $4.2984/M | 1M |
| Deepseek V4.1 Flash🇨🇦 | $0.2377/M | $0.9507/M | 1M |
| GLM 5.3🇨🇦 | $2.2183/M | $6.9718/M | 1M |
| GLM 5.3 Flash🇨🇦 | $0.1729/M | $0.5762/M | 1M |
| Kimi K3🇨🇦 | $5.6178/M | $28.0889/M | 1M |
Pricing shown in CAD. Billing by token. All prices subject to change — check docs for current rates.
What would you save? Let's do the math.
Pick the model your team burns today, drag the 30-day token budget, and see the same workload priced across our catalog — cheapest first, live rates, converted at today's Bank of Canada rate.
1.2 Savings calculator →
Monthly cost of 141M tokens on each Runars model
Cheapest on our catalog: GLM 5.3 Flash — you'd save $3,917 every month (99% less).
Assumes a 3:1 input:output split; real workloads vary. Your-model prices are USD list prices converted at the Bank of Canada rate of 1.3985 (2026-10-29). Runars rates are live CAD catalog prices — no FX line on your invoice. Monthly = blended rate × tokens.
The same prompt, again and again, for 10% of the price.
Mark a reused system prompt with an Anthropic-style cache_control block and Runars keeps it warm on Canadian soil. One request pays full price; every hit after it reads the cache.
1.0 Prompt Cache →
- Write1.25×–2×The first request writes your prompt to the cache in Montréal and pays a one-time premium.
- Warm5 minHeld warm for five minutes — or an hour — on Canadian soil. Your prompt never leaves the country.
- Read10%Every request inside the window reads the cache and bills at a tenth of the normal input price.
Built for any workload.
Inference, training, batch jobs, agents. If it runs in a container, it runs on Runars.
3.0 Use Cases →
Inference
Run LLM inference endpoints that scale with demand.
Fine-tuning
Train and fine-tune models on your data, your schedule.
AI Agents
Long-running autonomous tasks without server management.
Batch Processing
Compute-heavy workloads that run when needed, scale to zero.
Ready to run AI in Canada?
Start with the inference API for instant compatibility. Scale to GPU rental when you're ready. No lock-in, no setup fees.
npm install @anthropic-ai/sdk