If you already use Claude Code, Codex, or another terminal coding agent but do not want a separate model budget for every client, this setup is for you. AceDataCloud Coding Plan is a 30-day, usage-based API package: the current entry tier is Coding Plus at ¥99 (about $14.65) for 500 Credits, including a limited-time 100-Credit bonus scheduled to end November 30, 2026. It is not an official Claude or ChatGPT membership, and it is not unlimited.
The practical value is simpler: keep the client you know, choose a compatible protocol, and spend one Credits balance across available coding-model ecosystems.
Start with the protocol, not the model name
Coding clients do not all speak the same API protocol. Treating every client as “OpenAI-compatible” is a common source of failed integrations.
AceDataCloud exposes three useful paths for coding workflows:
| Client type | Protocol to select | Typical client |
|---|---|---|
| Anthropic-native agent | Claude Messages | Claude Code |
| OpenAI agent workflow | OpenAI Responses | Codex |
| Custom-provider client | OpenAI-compatible Chat Completions | Kimi Code, OpenCode, and other compatible clients |
The plan currently covers model ecosystems including Claude, GPT/Codex, GLM, Kimi, and DeepSeek. That does not mean every model works in every client. The client must support the model's protocol, and your Coding credential must expose that exact model ID.
Before editing any client configuration, query the live directory:
export ACEDATACLOUD_API_KEY="YOUR_CODING_APPLICATION_KEY"
curl https://api.acedata.cloud/v1/models \
-H "Authorization: Bearer $ACEDATACLOUD_API_KEY"
Copy an exact id from the response. Do not substitute a marketing name or an API UUID.
Minimal Claude Code configuration
Claude Code uses the Anthropic Messages path, so configure its base URL and authentication token rather than forcing it through a generic Chat Completions adapter.
For macOS, Linux, or WSL:
export ANTHROPIC_BASE_URL="https://api.acedata.cloud"
export ANTHROPIC_AUTH_TOKEN="YOUR_CODING_APPLICATION_KEY"
unset ANTHROPIC_API_KEY
Then start in a repository:
cd /path/to/your/project
claude
Use a small, read-only first task, such as:
Read README.md and summarize the local setup in five bullets. Do not modify files.
For project-local configuration, put the environment values in .claude/settings.local.json and add that file to .gitignore. Do not commit a real key to the shared .claude/settings.json.
Minimal Codex configuration
Codex uses the OpenAI Responses workflow. Keep the key in an environment variable:
export ACEDATACLOUD_API_KEY="YOUR_CODING_APPLICATION_KEY"
Add a provider to ~/.codex/config.toml:
model_provider = "acedatacloud"
[model_providers.acedatacloud]
name = "acedatacloud"
base_url = "https://api.acedata.cloud/v1"
[model_providers.acedatacloud.auth]
command = "sh"
args = ["-c", "echo $ACEDATACLOUD_API_KEY"]
Start Codex, run /model, verify that the provider is acedatacloud, and select one of the model IDs available to the credential. This is intentionally different from the Claude Code configuration: the shared host and credential do not erase protocol differences.
Divide models by task instead of changing tools
A multi-model plan is most useful when routing decisions are explicit. A simple policy is enough:
- Repository analysis and risky refactors: choose a stronger reasoning or code-understanding model, then require a written plan before edits.
- Unit tests, formatting, and repetitive fixes: use a lower-cost compatible model and keep the context limited to relevant files.
- A second opinion: send the patch and acceptance criteria to a different model family instead of repeating the same prompt in the same conversation.
- Documentation: provide the public API surface and examples, not the entire repository.
This policy improves budget control more than switching models at random. Credits are deducted according to uncached input, output, cache writes, and cache reads. Long context, tool calls, and parallel agents can therefore cost more even when the visible prompt is short.
Validate the integration before a real refactor
Use this sequence for any supported client:
- Confirm the key came from a Coding Application.
- Call
/v1/modelsand save an exact compatible ID. - Run a read-only request with a deterministic instruction such as “Reply only OK.”
- Run one bounded repository task without file writes.
- Check the request in AceDataCloud usage history and verify that Credits were deducted from the Coding Plan.
- Allow one small edit, review the diff, and run the project's own tests.
A 401 usually means the key is missing, invalid, or shadowed by an older environment variable. A “model not found” error means you should re-check the exact ID, credential scope, and client protocol. If a model appears in the general catalog but not in the Coding credential's live directory, do not assume the plan includes it.
Choose a tier from measured usage
All four tiers currently have the same Coding model access; they differ mainly in included Credits. During the current promotion they are:
| Plan | Credits | Display baseline | USD price |
|---|---|---|---|
| Coding Plus | 500 | ¥99 | $14.645677 |
| Coding Standard | 1,600 | ¥299 | $44.232902 |
| Coding Pro | 4,000 | ¥699 | $103.407354 |
| Coding Max | 10,000 | ¥1,399 | $206.962643 |
The Credits shown include the temporary bonus. Checkout is authoritative for eligibility and the final amount. Plans do not auto-renew and do not fall back to a general account balance. Renewing before expiry carries remaining Credits forward and extends the term by 30 days; otherwise unused Credits expire.
For a first integration, the smallest tier provides the same access scope with a lower fixed budget. Measure one week of your real tasks, record Credits by task category, and only then choose a larger tier.
The Coding Plan page contains current checkout information. The configuration center provides client-specific guides.
Top comments (0)