Groups

Understand how groups control API key access, routing, and pricing.

What groups do

Groups control which models, channels, and pricing rules an API key can use. Different groups can route the same model differently and may have different pricing.

Common groups

Choose one or more groups when you create an API key. Current user-visible groups include:

Group	Common use	Notes
auto	Default when you do not need fixed routing	Picks from active groups based on current routing rules. By default, prefers the pricier but more stable openai and anthropic groups for quick validation; switch to specific groups for long-term use
openai	OpenAI SDKs and OpenAI-compatible clients	GPT, Responses, and OpenAI-compatible image models
anthropic	Anthropic SDKs and Claude Messages clients	Claude Messages format
claude-bedrock	Claude Code and Claude clients	AWS Bedrock Claude routing; higher price, more stable, supports prompt caching
claude-kiro	Claude Code and Kiro-compatible use cases	Lower-cost Kiro-compatible Claude routing; may stop when quota runs out until an admin tops it up; channel limits can prevent cache hits
codex	Codex CLI	Gateway routing with compatibility priority
codex-passthrough	Codex CLI	Direct passthrough, lower latency, CLI-focused
codex-cheap	Codex CLI	Discounted gateway routing (expired on 2026-05-05; replacement pending)
codex-cheap-passthrough	Codex CLI	Discounted direct passthrough, CLI-focused (expired on 2026-05-05; replacement pending)
gemini	Gemini CLI and Google SDKs	Gemini API format
xai	Grok / xAI models	xAI routing
doubao	Doubao models	ByteDance model routing
opensource	Open-source models	Standard open-source model pricing
opensource-discount	Open-source models	Discounted open-source model pricing
opensource-economy	Open-source models	Economy open-source model pricing
opensource-highquality	Open-source models	High-quality open-source model routing

Group names, ratios, and available models follow the console and /api/pricing response. This table is a snapshot of user-visible groups and status, not a permanent price list.

Groups and model routing

AIOHub uses the API key group and requested model name to select a matching upstream channel. The same model can have different routing and pricing under different groups.

Recommended setup

• Create a separate API key for each client so you can track usage and limits by tool.
• If you are unsure which group to use, start with auto, then verify the target client, model, and endpoint with /v1/models and a small request.
• For Codex CLI, prefer codex or codex-passthrough instead of validating through a generic OpenAI group.
• Use the console or pricing page to check which models each group currently supports.

Notes

Claude Code 1M context

Some Claude models are natively configured for 1M context, such as claude-opus-4-7. These models do not need extra configuration for 1M context.

For models whose default context is not 1M, such as claude-opus-4-6, choose the opus[1m] model variant in Claude Code when you need the 1M context window. Without the [1m] suffix, Claude Code uses the default shorter context window.

Do not ask the model to confirm its own name or context length

A model’s answer to “what model are you” or “what is your context window” is not reliable. Model names and context limits are assigned after training; the model may not know its serving identifier or runtime limits. Use the client model selector or AIOHub console logs as the source of truth.

Kiro and PDF file support

Kiro does not natively support PDF input. When you ask Kiro to read a PDF, the client usually tries to call local tools to convert the PDF to text, similar to some desktop chat clients. If the local environment cannot convert the file, the request may fail.

If you need direct PDF handling in a conversation, use the claude-bedrock group. AWS Bedrock supports PDF files as native input without an extra local conversion step.