# Rate Limits and Throughput

This page covers RPM and TPM limits for [Sunrise Model as a Service](index.md) subscriptions. For an overview of the service, see the [MaaS landing page](index.md).

---

## Rate limits per subscription

Every MaaS subscription has **requests-per-minute (RPM)** and **tokens-per-minute (TPM)** limits. These apply **per subscription**, not per user or per API key, and are tied to your plan tier. All API keys created on the same subscription draw from the same RPM and TPM budget.

| Plan tier | Requests per minute (RPM) | Tokens per minute (TPM) |
|---|---|---|
| [Evaluation / trial](subscriptions/index.md#1-evaluation-plan) | 50 | 100,000 |
| CHF 1 to CHF 500/month | 200 | 1,000,000 |
| CHF 501 to CHF 1,000/month | 500 | 2,000,000 |
| CHF 1,001/month and above | [Contact sales for custom RPM/TPM](https://support.ai.sunrise.ch/) | — |

When you exceed either limit, further requests within that minute are rejected until the window resets. If you consistently hit these ceilings during normal use, review the guidance below or consider adding credits to move into a higher tier. See [Subscription plans](subscriptions/index.md).

!!!info Per subscription, not per key
Creating additional API keys does not increase your RPM or TPM allowance. Limits are enforced at the subscription level so that usage stays predictable across your organisation.
!!!

---

## Model context window vs. rate limits

A model's **context window** is separate from RPM and TPM. It is the maximum number of tokens a single request can contain (input plus output). Raising your TPM ceiling does **not** increase a model's context window.

---

## Why these limits exist

MaaS runs on shared infrastructure that serves many customers simultaneously. RPM and TPM limits keep the platform stable under load: without them, a single heavy consumer can consume disproportionate capacity and degrade response quality or availability for everyone else on the platform.

These limits exist to **ensure the stability of our MaaS offering for all customers**, not to restrict legitimate use. They are a fairness and capacity mechanism, distinct from your monthly credit allowance. You can still have credits available while temporarily hitting a per-minute throughput ceiling during burst traffic.

---

## Agentic workloads and TPM

TPM limits were rarely a practical concern until teams began using models such as GLM 5 for **agentic coding**. Agent workflows, including tool calls, MCP context, long conversation history, and repeated prompt expansion, consume tokens far faster than chat or document workloads. That adoption is a healthy signal, but it has changed the load profile on the platform faster than the underlying infrastructure was sized for.

If you are building coding agents or IDE integrations, plan for higher token throughput and consider models with larger context windows where appropriate. See [Active Models](active-models.md) for context window sizes and pricing.

---

## Higher throughput coming soon

Sunrise will soon roll out plans that let companies **purchase a larger reserved amount of TPM** on top of their existing subscription. This will give organisations with sustained high-throughput needs a dedicated capacity allocation without affecting other customers on the platform.

Until those plans are available, use the tier table above as your reference, and contact your Sunrise account team if you need guidance on sizing your subscription.

---

## Related links

[!ref icon="stack" text="Active Models: context windows, pricing, and model identifiers."](active-models.md)
[!ref icon="credit-card" text="Subscription plans and credit allowances."](subscriptions/index.md)
[!ref icon="pulse" text="Model health and availability on the status dashboard."](https://status.ai.sunrise.ch/)