Resource groups

Some models are served through multiple upstream vendor pools — for example a premium low-latency pool and a discounted throughput pool. Resource groups let you choose the pool per request; if you don’t choose, the platform’s default group applies and routing is fully automatic.

Selecting a group

Append the group name to the model with a colon:

{
  "model": "deepseek-ai/DeepSeek-V3.2:economy",
  "messages": [{ "role": "user", "content": "..." }]
}

Or set it in a header, leaving the model field untouched (useful when the model name is fixed by an SDK or framework):

curl https://inference.inferel.ai/v1/chat/completions \
  -H "Authorization: Bearer $INFEREL_API_KEY" \
  -H "X-Resource-Group: economy" \
  ...

A suffix is only interpreted as a resource group when it matches a group configured for your deployment — model names containing colons are otherwise passed through unchanged.

How routing works inside a group

Within the selected pool, requests are balanced across vendors by configured weight, scaled by live endpoint health, with KV-cache-aware stickiness for conversations that share a prefix. Saturated or unhealthy endpoints are avoided automatically.

Pricing

Each group carries a price multiplier applied to the model’s base per-1M-token rates (for example, an economy pool at 0.8× or a premium pool at 1.2×). The multiplied rate is what your credit balance is debited.

The set of available groups is configured per deployment. Contact support@inferel.ai to learn which pools are available for your organization.