LLM groups
Discover model-specific groups and prices, select a group through the model ID, and keep existing reAPI clients compatible.
One key, optional group selection
On the unified reAPI LLM endpoints, keep your normal reAPI key. A bare model ID, such as gpt-6-sol, uses its configured default. To choose a specific available group, use MODEL_ID:GROUP_ID as the model ID. No separate key or custom group header is required.
Groups belong to individual models. A group offered for Gemini is not automatically available for GPT or Claude. The discovery response is authoritative for the selected model; do not infer availability from a group name or another model's page.
Find groups on the model page
Open the model’s Pricing section to see its currently available groups, live prices and complete calling IDs. Copy the ID next to the group you want and use it in your existing request. For example, see GPT-6 Sol group pricing, or find another model in the catalog.
The Default column displays and copies the bare model ID: for example, claude-opus-5-5 continues to use its original claude group. Other columns copy the complete MODEL_ID:GROUP_ID. The Playground uses the same IDs. You do not need to call a discovery API manually to choose a group.
Programmatic discovery (optional)
GET https://reapi.ai/api/v1/models/MODEL_ID/groupsReplace MODEL_ID with the exact base API model ID from its model documentation. This read-only endpoint returns:
| Field | Meaning |
|---|---|
model | Base model ID queried |
defaultGroup | Configured default group, or null when none is available |
data[].id | Exact group ID |
data[].modelId | Bare model ID for Default; MODEL_ID:GROUP_ID for another group |
data[].available | Always true in the returned list; retained for older clients |
data[].ratio | Effective group multiplier |
data[].prices | Current valid per-million-token rates |
The list includes only groups authorized for the configured gateway account, with an enabled channel for this model and valid pricing. Use any returned entry’s exact modelId; no additional available filter is needed. If none qualify, data is empty. An unavailable configured default is reported as defaultGroup: null, without silently selecting a different group. Discovery does not reserve capacity: permissions, model availability and pricing are checked again on submission. Backend configuration supplies the groups and rates; model pages refresh automatically. There is no two-group or five-group API limit.
OpenAI and Anthropic requests
For /api/v1/chat/completions, put the full ID in the usual model field. For Claude models that support /api/v1/messages, use the same ID format in the native Messages body. Keep the remaining fields in the format documented for that endpoint.
{
"model": "gpt-6-sol:GROUP_ID",
"messages": [{"role": "user", "content": "Hello"}]
}GROUP_ID is a placeholder; replace it with an ID discovered for this exact model. Streaming and non-streaming requests use the same selection.
Native Gemini requests
Google's GenerateContent API defines generateContent for a complete JSON response and streamGenerateContent for SSE streaming. These are API methods, not model names. See Google's API reference.
POST /api/v1beta/models/MODEL_ID:generateContent
POST /api/v1beta/models/MODEL_ID:streamGenerateContent?alt=sseOn reAPI, MODEL_ID can be the base ID or a discovered MODEL_ID:GROUP_ID. For example:
POST /api/v1beta/models/gemini-4-hydrogen:GROUP_ID:generateContentThe group suffix is a reAPI routing extension, not part of Google's official model ID. The last colon separates the method from the model ID. URI-encode the complete model ID as one URL path segment when needed; keep JSON model IDs as literal strings. Native request bodies, including signed parts and unknown extension fields, retain their existing format.
Downstream NewAPI channels
For an OpenAI-compatible NewAPI channel, use https://reapi.ai/api as the base URL and your normal reAPI key as the channel key. Add the exact calling ID copied from the price table: the bare model ID for Default, or MODEL_ID:GROUP_ID for another group. NewAPI sends it through the ordinary model field; no custom group-header configuration is needed. OpenAI SDK clients use the base URL https://reapi.ai/api/v1.
Existing clients and billing
- Existing bare model IDs, authentication methods and endpoint paths remain valid. They do not require a group lookup or code change.
- Existing
X-ReAPI-Groupclients remain supported. If both the suffix and a non-autoheader select different groups, the request fails with 400 before generation. - Legacy headers with non-ASCII names still use URI encoding plus
X-ReAPI-Group-Encoding: uri. New integrations can use the model ID instead. - An unavailable or unauthorized explicit group fails before generation; it does not fall back to another group with another price. Fixed-group gateway keys remain restricted to their group.
- Charges follow the selected group's settled request price. The group multiplier is not applied twice. This selection adds no automatic retry or extra generation.
- Response model identifiers retain the base model name. Group selection is recorded separately for billing; the suffix is not sent to the model provider.
Models that already used NewAPI before group selection retain their original credential's group for bare-model calls. Adding groups, changing their prices or reordering AutoGroups does not replace that route. If the original group becomes unavailable, the call fails instead of switching groups. Models newly migrated from another service use their configured NewAPI default.
The separate legacy api.reapi.ai workspace uses its existing keys and routing rules. This guide does not migrate its models or change its API contract. Follow each model page's service-status note.