Model setup

Use qwen-plus in DeepSeek Harness

Connect qwen-plus with your ChineseModels API key and keep other providers in place.

Set up inCodex Kimi Code DeepSeek Harness Claude Code Hermes Agent OpenClaw DeepSeek guide

Copy the setup prompt

Paste into your coding assistant; enter your key only through a local secure prompt or secret store.

Set up qwen-plus in DeepSeek Harness

Configure qwen-plus from ChineseModels in DeepSeek Harness.
Guide: https://chinesemodels.ai/documentation/deepseek-harness/models/qwen-plus
Exact model ID: qwen-plus

Detect my OS and client version; check the current official configuration reference. Back up existing config and preserve logins, providers, permissions and unrelated settings. Use a secure local key prompt or secret store; never print or commit the key.

Use Settings > Models > Add model provider > Custom model API. Set provider ID chinesemodels, base URL https://chinesemodels.ai/v1 and OpenAI Chat Completions. Enter the credential locally, fetch available models, select text models and save. Preserve other providers in cordis.patch.yml. For DeepSeek V4, verify compat.thinkingFormat: deepseek and the supported reasoningEfforts; do not apply this mapping to other families.

Fetch https://chinesemodels.ai/v1/models with my bearer key. Confirm this exact ID exists; do not silently substitute another model. Exclude image/video generators when importing more models. Test text, streaming, a tool call and a tool-result round trip. Only configure verified thinking parameters. Report what was tested, how to select this model, refresh the list and restore the backup. Do not change messaging channels, send messages or publish anything.

Manual setup

Start DeepSeek Harness, then open Settings > Models > Add model provider > Custom model API.
Set provider ID chinesemodels, base URL https://chinesemodels.ai/v1 and OpenAI Chat Completions. Enter your key in the credential field.
Fetch available models, select text models and save. You can add the exact ID manually if discovery fails.
Select qwen-plus in the picker and start a fresh session.

Start the official Web UI with Node.js installed

npx @deepseek-ai/dsh web

Configuration fragment: merge, do not replace the whole file

- id: llm-pi-ai
  config:
    providers:
      chinesemodels:
        apiKeyEnv: CHINESEMODELS_API_KEY
        api: openai-completions
        baseURL: https://chinesemodels.ai/v1
        compat:
          supportsDeveloperRole: false
          maxTokensField: max_tokens
        models:
          - id: "qwen-plus"
            input: [text]

Advanced settings for the default Web UI profile live in $DSH_HOME/profiles/web/cordis.patch.yml. Preserve the full existing plugin entry when merging.

Thinking and validation

Start with provider defaults. Only enable reasoning controls whose request fields match the chosen model. A visible effort menu does not establish that upstream thinking can be switched off.

Refresh from your authenticated /v1/models list and exclude image/video generators. Test a text reply, streaming and a tool-result round trip before using a model for agent work. Restore your configuration backup to undo the setup.

These additional client guides follow official configuration documentation. Complete end-to-end tests have not been performed for every client/model combination. Codex and Kimi guides document their own test results.

Official DeepSeek Harness configuration reference