#Core Concepts

This page explains ByteSeek's basic model: groups, pools, multipliers, balance and quota — and the gates a request passes through before it is allowed (or rejected).

#Glossary

#Groups

The group is the most important concept — it determines four things at once:

1. Which models you can use — the model must be in the group's list;
2. At what price — base price × group multiplier;
3. Which protocol format is accepted — OpenAI, Anthropic and more (full list in API Endpoints), and it must match your client;
4. Which pool serves you — determining capacity and peak-hour behaviour.

Creating an API key requires picking a group. To use several groups with one key, see Composite Keys.

#Format groups

Some models are split into two groups by protocol format, e.g. "DeepSeek (OpenAI format)" and "DeepSeek (Anthropic format)". Choosing the wrong format gets you a 403 — confirm what protocol your client speaks first, see API Endpoints.

#Client-restricted groups

Some groups only accept specific clients (e.g. Codex-only, Claude-Code-only); other clients receive 403. Every group in the model plaza carries an explanatory label — similarly named groups can have completely different restrictions.

SCREENSHOT

#Group ≠ subscription plan

A group is a property of an API key; a subscription plan is a payment method for the account. Names may look similar, but they are unrelated: quota from any plan can be spent on any group you have access to.

#Pools

A pool is the set of upstream accounts behind a group. One pool can serve several groups, each with different multipliers and client restrictions. Pool capacity and health determine peak-hour queueing; see each group's notes in the model plaza.

#API keys and scope

Scope is fixed at creation time and toggled in the top-right corner of the API Keys page:

• A regular key binds one group; write the model ID directly when calling.
• A composite key binds several groups; write prefix/modelID, see Composite Keys.

#Cost calculation

A request is priced in three parts — input, output, cache:

cost = input tokens × input price + output tokens × output price + cache tokens × cache price
(each price = model base price × group multiplier)

• Cache hits are billed at the cache price, usually far below the input price;
• Composite keys are billed by the group the prefix actually hits;
• Fast Mode uses the high-priority lane at roughly 2× the regular price.

Deduction order: subscription first, then balance. The subscription expiring soonest is consumed first; once subscriptions are exhausted, the balance takes over. Details in Billing.

#The five gates of a request

Every request passes these gates in order — check them in the same order when troubleshooting:

Status codes and messages for each gate are in Error Codes.

#Related

• Create an API Key — pick a group, generate a key
• Billing — multipliers, subscriptions and balance
• API Endpoints — the URLs your client needs
• Troubleshooting — locate failures by symptom