If you use Claude Code, Codex, or an OpenAI-compatible coding client and want one usage budget instead of separate subscriptions, this guide shows how to choose the right protocol, configure the smallest useful request, and verify that billing is going to the intended Ace Data Cloud Coding Plan.
The current public catalog starts with Coding Plus: 500 Credits for 30 days, priced at $14.645677 with a ¥99 RMB display anchor. The 500-Credit amount includes a 100-Credit limited-time bonus scheduled through November 30, 2026; checkout remains authoritative for eligibility. Credits are usage-metered rather than “unlimited,” and every tier exposes the same Coding model access.
Start with the protocol, not the model name
A common integration mistake is treating every model as if it speaks the same API. A client may support a model family but still require a specific wire protocol.
The Coding Plan currently publishes three useful protocol paths:
| Workload or client | Protocol to select | Public endpoint |
|---|---|---|
| Claude-native agents and Claude Code-style integrations | Anthropic Messages | /v1/messages |
| Codex-style agents | OpenAI Responses | /openai/responses |
| General multi-model clients | OpenAI-compatible Chat Completions | Family-specific paths such as /openai/chat/completions, /glm/chat/completions, /kimi/chat/completions, or /deepseek/chat/completions
|
This distinction matters because a valid token does not make an incompatible request shape valid. A Responses client expects Responses semantics; a Claude-native client expects Messages content blocks and Anthropic-style fields. Do not solve a protocol error by randomly changing model IDs.
The public service catalog currently includes Claude, GPT/Codex, GLM, Kimi, and DeepSeek options. Exact IDs can change, so copy the model ID shown by your Coding credential’s live directory or the relevant client profile instead of hard-coding a name from an old tutorial.
Minimal configuration
Create a Coding Application in the Coding Plan product page, then copy its token. Use placeholders in committed configuration:
export ACE_CODING_TOKEN="replace-with-your-coding-application-token"
export ACE_API_BASE="https://api.acedata.cloud"
export MODEL_ID="copy-an-exact-compatible-model-id"
Keep the token outside source control. For a GUI or coding client, map these same values into its base URL, API key/token, and model fields, but choose the client profile that matches the protocol table above. The configuration center contains client-specific profiles for tools including Claude Code, Codex, Kimi Code, OpenClaw, and others.
Do not assume that every client can use every model. Compatibility is the intersection of:
- the client’s supported protocol;
- the endpoint selected in its profile;
- the current model directory; and
- the permissions of the Coding credential.
Verify authentication without generating an answer
For a Claude Messages profile, the public catalog exposes a token-count endpoint. It is a useful first check because it validates the base URL, bearer token, request format, and model ID without asking the model to generate output:
curl --fail-with-body \
"$ACE_API_BASE/v1/messages/count_tokens" \
-H "Authorization: Bearer $ACE_CODING_TOKEN" \
-H "Content-Type: application/json" \
-d "{
\"model\": \"$MODEL_ID\",
\"messages\": [
{\"role\": \"user\", \"content\": \"Return a one-line summary.\"}
]
}"
A successful response contains an input_tokens integer. This verifies the connection, not response generation quality. If it fails, debug in this order:
- 401: token missing, malformed, or no longer valid;
- 403: credential permission or available-plan balance problem;
- 404: wrong path or API base;
- 400: request schema or model ID mismatch;
- 429: rate limit reached; back off rather than immediately multiplying retries.
After that, run one small generation request through the client you actually intend to use. Confirm that the client launches with the expected profile and that the Coding Plan usage view records the call.
Assign models by task instead of defaulting to one
The practical value of a multi-model plan is routing work deliberately:
- Use a lower-cost compatible model for repository discovery, formatting, test scaffolding, and repetitive edits.
- Move architecture decisions, difficult debugging, or large refactors to a stronger reasoning model.
- Keep a second family available when a task benefits from a different tool-calling or coding style.
- Pin a model ID in CI and unattended jobs; interactive experimentation and automation need different change-control policies.
This is more predictable than selecting the most capable model globally. Long context, parallel agents, tool calls, output length, and cache behavior all affect consumption.
Choose a tier from a workload budget
The public catalog currently lists these 30-day packages:
| Tier | Current promotional Credits | USD price | RMB display anchor |
|---|---|---|---|
| Coding Plus | 500 | $14.645677 | ¥99 |
| Coding Standard | 1,600 | $44.232902 | ¥299 |
| Coding Pro | 4,000 | $103.407354 | ¥699 |
| Coding Max | 10,000 | $206.962643 | ¥1,399 |
Do not mix the USD price field with the RMB cny_anchor; they are different currencies and serve different display purposes. For rough internal budgeting, the effective promotional USD per Credit is price / amount. That is about $0.02929/Credit for Plus and $0.02070/Credit for Max, but actual task cost still depends on the selected model and metered token categories.
The plan has no automatic renewal and does not fall back to a general account balance. Renewing before expiry carries remaining Credits forward and extends the term by 30 days; unused Credits expire when the term ends, and a later purchase starts a new term with only the new package amount.
A safe rollout checklist
- Buy the smallest tier that covers the next 30 days, not an imagined annual workload.
- Create a dedicated Coding credential and keep it out of Git.
- Match the client to Messages, Responses, or Chat Completions.
- Copy an exact current model ID.
- Run a non-generating or minimal request.
- Check the usage record before enabling long context or parallel agents.
- Set application-side retry limits and a monthly or per-job Credit budget.
For source details, see the Coding Plan catalog, integration guide, and Ace Data Cloud’s articles on choosing the ¥99 plan and consolidating coding clients.
Top comments (0)