Sovereign access to the best open models.
One subscription to top open-weight models, hosted exclusively on European infrastructure by European-owned providers. Zero data retention, GDPR-native, and it works with any agent you already use.
- GLM-5.2
- Mistral Large 3
- Qwen3.6
- Qwen3.5
- Devstral 2
- Mistral Small 4
- DeepSeek V4Soon
- Kimi K2.7 CodeSoon
- MiniMax M3Soon
One flat subscription. No per-token anxiety.
A price you can plan around, with limits stated up front instead of a meter running behind your back.
Arc
50% off first monthThe full catalog: every open model we host, including the frontier tier.
then €10/month · launch offer
- Every model in the Arc catalog, nothing gated, GLM-5.2 included
- Up to ~1,000 requests per 5-hour window on efficient coding models
- Generous rolling limits: per 5 hours, per week, per month
- Works with Radian and any agent you already use
- Cancel any time
Arc+
50% off first monthThe same catalog with real headroom: higher limits in every window.
then €20/month · launch offer
- Same models as Arc: the catalog is never split
- Higher limits across every window: per 5 hours, per week, per month
- Room for frontier-heavy, all-day agent workflows
- Works with Radian and any agent you already use
- Cancel any time
Prices include VAT. The first-month discount is a launch offer. Request estimates assume typical agent requests; actual counts depend on model and context size.
Transparent limits, not fine print.
Every plan is metered across three rolling windows. Arc+ carries a substantially larger allowance in each one. As a feel for scale: Arc covers up to ~1,000 requests per 5-hour window on efficient coding models.
Per 5-hour window
A rolling window, so a burst of agent work never drains the month.
Per week
The weekly ceiling resets on a rolling basis, not on a billing date.
Per month
The outer bound. Hit it and you wait or top up, never a surprise bill.
Bars show relative allowance per plan, not absolute request counts. Request estimates assume typical agent requests; actual counts depend on model and context size. Per-model estimates ship with the product.
Your tokens shouldn't emigrate.
Open-weight models closed the capability gap. What's missing is a way to use them without your prompts crossing a jurisdiction you didn't choose. Arc is that way.
A sovereign alternative
Most inference today routes through US-owned providers, wherever the GPUs sit. Arc serves every token through European-owned providers, not subject to the US CLOUD Act.
Zero data retention
Your prompts and outputs are never stored, never used for training, and never leave the EU, transient operational data excepted.
No training on your work
Nothing you send through Arc trains anyone's model. Full stop.
EU hosting, end to end
Models are hosted, served, and metered inside EU data centers. GDPR-native processing isn't a compliance add-on; it's where the traffic lives.
A price you can plan around
A flat monthly subscription with transparent rolling windows, instead of per-token metering you audit at the end of the month.
Works with any agent
Arc is a gateway, not a walled garden. Point Radian at it, or any agent or tool you already use.
Three steps, then it's boring.
A gateway should disappear into your setup. Arc takes minutes to adopt and nothing to maintain.
Create your account
One Radian account covers Arc. No sales call, no procurement form: sign up and you're in.
Pick Arc or Arc+
€10 or €20 a month, half off your first month. Cancel any time; the subscription is the only contract.
Point your agent at Arc
Drop the Arc endpoint and your key into Radian, or into any agent or tool you already use, and keep working. The models change; your workflow doesn't.
Frequently asked questions.
Straight answers about what Arc serves, where your prompts go, and what the subscription actually covers.
Keep your tokens on European soil.
Arc opens with the beta. People on this list get access, and the 50%-off first month, before anyone else.
Arc €10 · Arc+ €20 · half off your first month