Rate limits

Rate limits apply per API key — RPM, TPM and concurrency — not to your total token volume.

Each API key has a limit of 2 million tokens per minute (2M TPM), including keys enabled for the GLM 5.3 add-on. Monthly token allowances depend on the plan and model. Monthly quota expansion is not available.

Per-key limits

Consolidated keys multiply every limit above by the number of seats on the key.

Embedding and speech models have their own request limits. See Models. GLM 5.3 is an add-on with its own limits and its own allowance, further down this page.

Monthly volume

Efficient open models are unlimited on every plan. The frontier-class models — DeepSeek V4.1 Flash and GLM 5.3 Flash — count against a monthly quota that scales with your plan:

GLM 5.3 (the add-on) carries its own separate allowance, further down this page; the two quotas never draw from each other.

Monthly quota expansion is not available. See Pricing for the full breakdown.

Images

Image generation has its own monthly allowance, per organisation, and it does not draw from the token quota above:

Two things decide how fast that runs down:

• The cap counts images produced, not requests. One call asking for n: 4 spends four.
• Deleting an image does not refund it. The month’s usage is what was generated, not what you kept.

One bucket per organisation, shared by every surface: images made in the console and images made through /v1/images/* come out of the same total.

GLM 5.3

GLM 5.3 is a per-key add-on at €150 a month, and no plan covers it. It carries its own per-key limits:

Its monthly allowance belongs to your organisation and is shared across every key with GLM 5.3. It is separate from the quota above and never draws from it.

Monthly quota expansion is not available. See Models.

Handling 429

When you exceed a limit, the API returns 429 Too Many Requests. Back off and retry. Most OpenAI SDKs already do this for you with exponential backoff — the Retry-After header tells you how long to wait.

# the OpenAI SDK retries 429s automatically — tune the count if needed
client = OpenAI(
    base_url="https://api.helmcode.com/v1",
    api_key="sk-your-key-here",
    max_retries=5,
)

Need higher limits?

High-demand workloads can get higher RPM and pooled concurrency on request. Dedicated and Enterprise plans also run on hardware reserved for you, with guaranteed throughput. Talk to us.