OpenClaw's Tencent Cloud provider defaults to hy4-preview. To explicitly use the GA release, set the primary model to tencent-tokenhub/hy4 or tencent-tokenplan/hy4, and make sure the API key for the selected endpoint has Hy4 access.
Suitable tasks: Calling Hy4 through Tencent Cloud TokenHub or TokenPlan in OpenClaw for text, images, long contexts, code, and agent workflows.
Unsuitable tasks: Tasks that require the model to generate images directly or take audio or video input; the documentation lists image input and text output only.
Applicable model version: The hy4 GA release. The documentation does not specify minimum versions for OpenClaw or the provider plugin.
Applicable client, agent, or API: OpenClaw; Tencent's official provider plugin, @openclaw/tencent-provider; the OpenAI-compatible openai-completions API.
Recommended reasoning levels and parameters: reasoning_effort accepts none or high; other reasoning levels map to high, which is also the default.
The two endpoints' provider IDs, key variables, and API addresses are as follows. Even after configuring authentication, the model ref must still explicitly point to hy4:
| Endpoint | Provider ID | Model ref | API key environment variable | Default API address |
|---|---|---|---|---|
| TokenHub | tencent-tokenhub | tencent-tokenhub/hy4 | TOKENHUB_API_KEY | https://tokenhub.tencentmaas.com/v1 |
| TokenPlan | tencent-tokenplan | tencent-tokenplan/hy4 | TOKENPLAN_API_KEY | https://api.lkeap.cloud.tencent.com/plan/v3 |
Interactive authentication:
openclaw onboard --auth-choice tokenhub-api-key
# 或
openclaw onboard --auth-choice tokenplan-api-keyNon-interactive authentication:
openclaw onboard --non-interactive \
--mode local \
--auth-choice tokenhub-api-key \
--tokenhub-api-key "$TOKENHUB_API_KEY" \
--skip-health \
--accept-risk
openclaw onboard --non-interactive \
--mode local \
--auth-choice tokenplan-api-key \
--tokenplan-api-key "$TOKENPLAN_API_KEY" \
--skip-health \
--accept-riskNon-interactive mode requires --accept-risk. Authenticate on the host running the Gateway. --mode remote only configures the local client connection; it does not configure provider credentials on the server.
In the OpenClaw primary model configuration, set agents.defaults.model.primary to the model ref for the endpoint in use. Select only one at a time:
{
"agents": {
"defaults": {
"model": {
"primary": "tencent-tokenhub/hy4"
}
}
}
}For TokenPlan, replace the value with tencent-tokenplan/hy4. hy4-preview is the default model for new configurations; completing provider authentication alone does not mean the current primary model has switched to hy4.
Check which models are visible for each endpoint:
openclaw models list --provider tencent-tokenhub
openclaw models list --provider tencent-tokenplan| Item | Hy4 GA |
|---|---|
| Input | Text, images |
| Supported image formats | JPEG, PNG, WebP, GIF |
| Output | Text |
| Context window | 1,024,000 tokens |
| Maximum input | 960,000 tokens |
| Maximum output | 64,000 tokens |
| Reasoning parameter | reasoning_effort: none or high; other levels map to high |
Restricted API keys must allow the hy4 model on the selected endpoint. The documentation recommends confirming model access before migrating to Hy4 and noting the pricing difference between Hy4 and older models.
Create an API key in TokenHub or TokenPlan. A restricted key must include hy4 access for the selected endpoint.
Choose the corresponding endpoint and configure authentication as described above.
Set agents.defaults.model.primary to the corresponding .../hy4 model ref.
Run openclaw models list --provider <提供方 ID> to check the model catalog, then make a request to verify the actual call.
If the Gateway is managed by launchd, systemd, or Docker, put the API key in ~/.openclaw/.env or provide it to the process through env.shellEnv. Variables exported only in an interactive shell are not visible to managed processes.
The documentation lists tencent-tokenhub/hy4 and tencent-tokenplan/hy4 as the GA model refs; the default ref for new configurations on both endpoints is .../hy4-preview.
The official plugin package is @openclaw/tencent-provider; both endpoints use OpenAI-compatible APIs.
The global TokenHub address override is https://tokenhub-intl.tencentmaas.com/v1. TokenHub also supports address overrides by account or region.
The built-in catalog lists Hy4 with a 1,024,000-token context window, a 64,000-token maximum output, and support for text and images; the documentation separately gives an input limit of 960,000 tokens.
The page does not publish minimum version numbers for OpenClaw or the plugin, or provide a plugin installation command.
openclaw doctor --fix does not automatically migrate an existing hy3-preview primary model to Hy4. When the TokenHub model allowlist is enabled, an existing hy3-preview primary model is migrated to hy3; explicitly configured hy3 and hy4-preview remain unchanged.
The documentation recommends explicitly choosing Hy4 for migration because its pricing differs and an account's model permissions may vary by endpoint.
The primary-model field example in this article sets the exact model ref listed in the documentation as the default model; the provider page does not list a separate Hy4 installation or version-pinning command.
Configure an API key for one endpoint as described in the source, set the primary model to that endpoint's hy4 model ref, confirm the Gateway process can read the key, then make a call and verify the response. Do not mix the TokenHub and TokenPlan model refs or API keys.
Hy4