One-Time Agent Plan

Agent Plan

Review the stocked one-time offer, compare supported models and GLM-based token capacity, then copy the correct configuration for your coding agent.

Current one-time offer

Agent Plan

Upstream key type: Volcengine Ark Agent Plan Medium. This page explains how to call the route directly and how to connect the supported coding tools.

Compare models & AFP View $59.90 package

Quota reference

GLM 5.2

Approximate usage

≈30M / 5H · ≈100M / 1 Week · ≈300M / 1 Month

Default config Model ID

glm-5.2

These buyer-facing estimates use GLM 5.2 as the reference model and its standard AFP coefficient of 4.5. Other models consume the shared plan capacity at their published AFP coefficients.

Quota windows refresh upstream: the 5-hour window rolls from your first request, the weekly quota resets every Monday at 00:00, and the monthly quota resets on the plan's monthly renewal date (provider time). The underlying Medium plan runs on one-calendar-month validity periods.

DeepSeek V4 ProDeepSeek V4 FlashGLM 5.2MiniMax M3Kimi K3

1 · Quick start

Start by calling the model

The key is a direct access credential. Put it in the tool or environment variable required by the client, then call the provider route with the default configuration Model ID below.

Get the key

Copy the delivered key from Dashboard. Do not place the raw key in a public repository or commit it into a configuration file.

Choose the protocol

Use the Anthropic-compatible route for Claude Code. Use the /v3 OpenAI-compatible route for Codex, OpenCode, OpenClaw, and Hermes.

Verify the route

Put the exact Model ID in the client configuration, save it, restart the tool, and confirm that the selected model responds normally.

2 · Model call

Test the OpenAI-compatible connection

This test confirms the key, base URL, protocol, and selected Model ID before you add a larger coding-agent workflow.

OpenAI protocol only

Both endpoints below use the /v3 OpenAI-compatible base URL. Claude Code uses a different protocol and is configured separately in the next section.

Chat Completions

https://ark.cn-beijing.volces.com/api/plan/v3/chat/completions

Responses API

https://ark.cn-beijing.volces.com/api/plan/v3/responses

export ARK_API_KEY="YOUR_AGENT_PLAN_KEY"

curl https://ark.cn-beijing.volces.com/api/plan/v3/chat/completions \
  -H "Authorization: Bearer $ARK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "glm-5.2",
    "messages": [{"role": "user", "content": "Reply with a short connection confirmation."}]
  }'

Replace the value of model with another exact ID from the supported-model table below when you want to change models.

3 · Model configuration and AFP

Choose the model in the configuration file

Choose the protocol for your tool first, then select a Model ID. The Model IDs are shared, but the base URL, key field, and request format are not interchangeable. The quota shown on the product card uses the standard GLM 5.2 coefficient of 4.5 as its reference conversion. Other models consume the shared capacity at the relative rates shown below.

Protocol A · OpenAI-compatible

Codex and OpenAI-style tools

Use with: Codex · OpenCode · OpenClaw · Hermes Agent
Base URL · includes /v3: https://ark.cn-beijing.volces.com/api/plan/v3
Key field: ARK_API_KEY
Request format: Responses or Chat Completions

Protocol B · Anthropic-compatible

Claude Code only

Use with: Claude Code
Base URL · no /v3: https://ark.cn-beijing.volces.com/api/plan
Key field: ANTHROPIC_AUTH_TOKEN
Request format: Anthropic Messages API

Do not mix the two routes. OpenAI-compatible tools need the /v3 URL and their OpenAI-style configuration. Claude Code needs the base URL without /v3 and the ANTHROPIC_* fields.

Model	Configuration Model ID	Current AFP coefficient	Tokens vs GLM 5.2
GLM 5.2 Default configuration and usage-reference model. Its standard AFP coefficient is 4.5.	`glm-5.2`	4.5	1× · baseline
DeepSeek V4 Flash Lowest AFP coefficient among the listed models, suited to higher-volume coding work.	`deepseek-v4-flash`	0.5	≈9×
MiniMax M3 A middle-consumption option between V4 Flash and the higher-coefficient routes.	`minimax-m3`	2.5	≈1.8×
Kimi K3 Kimi K3 provides a 1M context window for long coding and agent work; its 10 AFP coefficient yields about 0.45× the token capacity of the standard GLM 5.2 baseline.	`kimi-k3`	10	≈0.45×
DeepSeek V4 Pro The V4 Pro route for users who explicitly choose the higher AFP coefficient.	`deepseek-v4-pro`	5.5	≈0.82×

AFP formula

(input tokens × input coefficient + output tokens × output coefficient) ÷ 10,000

The input coefficient is the model coefficient multiplied by 0.67 for input up to 32K, 1 for 32K–128K, or 2 above 128K. Output uses the model coefficient directly. GLM 5.2 uses its standard coefficient of 4.5.

Official supported models & limits Official AFP rules

Anthropic protocol example · Claude Code

~/.claude/settings.json

{
  "env": {
    "ANTHROPIC_AUTH_TOKEN": "YOUR_AGENT_PLAN_KEY",
    "ANTHROPIC_BASE_URL": "https://ark.cn-beijing.volces.com/api/plan",
    "ANTHROPIC_MODEL": "glm-5.2",
    "ANTHROPIC_DEFAULT_HAIKU_MODEL": "glm-5.2",
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "glm-5.2",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "glm-5.2",
    "CLAUDE_CODE_SUBAGENT_MODEL": "glm-5.2"
  }
}

This JSON is only for the Anthropic-compatible Claude Code route. To change models, replace all five model values with the same selected ID. Do not copy these ANTHROPIC_* fields into an OpenAI-compatible client.

4 · Connect AI tools

Choose the access method for your tool

Each tool has its own protocol, secret name, and configuration file. Start with the matching card; do not copy the Claude Code route into an OpenAI-compatible client.

Codex

OpenAI Responses API

Configuration file: ~/.codex/config.toml
Secret / key location: ARK_API_KEY
Start here: Set ARK_API_KEY, then set the provider base URL and wire_api = responses.
Model fields to change: model = "glm-5.2"

Open our Codex reference

Claude Code

Anthropic-compatible API

Configuration file: ~/.claude/settings.json
Secret / key location: ANTHROPIC_AUTH_TOKEN
Start here: Set the Anthropic token and base URL, then use the same exact Model ID in every model field.
Model fields to change: ANTHROPIC_MODEL + HAIKU / SONNET / OPUS defaults + CLAUDE_CODE_SUBAGENT_MODEL

Open our Claude Code reference

OpenCode

OpenAI Responses API

Configuration file: ~/.config/opencode/opencode.json
Secret / key location: API key in provider.options
Start here: Create a provider with @ai-sdk/openai, the /v3 base URL, and the selected Model ID as both the registry key and the name field.
Model fields to change: model + provider.models["glm-5.2"]

Open our OpenCode reference

OpenClaw

OpenAI Responses API

Configuration file: ~/.openclaw/openclaw.json
Secret / key location: apiKey in models.providers
Start here: Merge the provider, agents, and gateway blocks, set the openai-responses adapter, and use the exact Model ID in the provider and primary route.
Model fields to change: models.providers[].models[].id + agents.defaults.model.primary

Open our OpenClaw reference

Hermes Agent

OpenAI Responses API (api_mode: codex_responses)

Configuration file: ~/.hermes/config.yaml
Secret / key location: model.api_key in config.yaml
Start here: Put provider, base URL, key, Model ID, and api_mode under the model block, then restart Hermes.
Model fields to change: model.default

Open our Hermes Agent reference

5 · Verify the configured model

Confirm the connection before starting a full project

Run one short request and confirm a normal response.

Confirm the configuration contains the exact selected Model ID in every required field.

Restart the tool and run its status or verification command again.

Read compatibility FAQ Open Dashboard