QwenCloud Token Plan: Lite vs Standard vs Pro, and Hermes Agent Setup

Quick decision: start with Lite ($6/mo) for one interactive Qwen session at a time; Standard ($18/mo) for most-day personal coding and agent work; Pro ($68/mo) for several concurrent agents. Not for production: Token Plan Individual is for personal interactive use only — automated scripts, batch jobs, and background scheduled tasks violate the terms, and the 5-hour and 7-day Credit windows pause the service mid-run.

Last verified: August 2, 2026, against the QwenCloud pricing page, the Token Plan docs, and the Hermes Agent providers documentation. Reference links are in the sources list at the end.

Firsthand status: Alastair Fraser has not personally used QwenCloud Token Plan. This is an independent research and setup guide based on current official documentation and public plan information; it does not claim an authenticated hands-on test.

QwenCloud Token Plan is a monthly, Credits-based AI subscription covering text, vision, image, and video models plus Harness tools. Lite / Standard / Pro cost $6 / $18 / $68 per month (limited-time), each with a 5-hour Credit cap and a 7-day Credit cap: every call deducts from both windows at once, and when either is exhausted the service pauses until that window elapses. Unused Credits do not carry over.

The detail that matters for Hermes Agent: Token Plan uses its own API key and its own base URL, separate from the pay-as-you-go DashScope key and the Coding Plan key. Mixing them produces the 401 errors at the top of QwenCloud’s own troubleshooting table, so this guide pairs them exactly as documented.

Three quick recommendations:

1. Start with Lite — enough to evaluate whether the Credit model fits your working rhythm.
2. Consider Standard — the practical everyday tier for daily Hermes use.
3. Skip Token Plan for production — the terms disallow automated and scheduled use; evaluate QwenCloud’s pay-as-you-go separately.

Jump straight to the setup: Five-step Hermes quick setup.

Contents

1. Short answer and recommendation
2. Plan comparison: Lite vs Standard vs Pro
3. Which plan should you choose?
4. Five-step Hermes quick setup
5. How the quotas actually work
6. Models, media, and exclusions
7. Command map
8. Troubleshooting
9. Security and billing

Short answer and recommendation

Token Plan is a Credits subscription for individual developers, sold in three tiers plus optional Credit Packs, covering text generation, reasoning, visual understanding, image generation, video generation, and speech synthesis from Qwen, Zhipu (GLM), DeepSeek, Wan, and HappyHorse models. It works with mainstream coding and agent tools — including Hermes Agent.

The numbers that govern usage are the 5-hour window (700 / 3,000 / 12,000 Credits) and the 7-day window (2,500 / 10,000 / 40,000 Credits). Credit Packs add 20,000 Credits per pack that are not subject to either window.

If you drive Hermes interactively and can live with the interactive-only terms, Standard is the sweet spot; Lite is a fine trial; Pro earns its price only with genuinely parallel agent work. For automated, scheduled, or customer-facing workloads, the terms do not permit Token Plan use — evaluate pay-as-you-go separately.

Plan comparison: Lite vs Standard vs Pro

The Credits are not a token bank. The 2,500 / 10,000 / 40,000 figures are Credit budgets, not Qwen tokens. Each call deducts Credits by per-model coefficients depending on model, modality, and (for video) duration and resolution. A short text exchange costs a small fraction of a Credit; a video call can cost tens and settles only after the task completes. The console’s usage page is the source of truth at call time.

The pricing page lets you pick monthly, quarterly, or yearly billing at checkout — treat the live checkout as authoritative, because the limited-time figures can change without notice. A separate Team Edition (per-seat monthly quotas, no 5-hour/7-day windows) exists for teams and is out of scope here.

Which plan should you choose?

• Lite: one active session, prototyping, cheapest entry. The 700-Credit 5-hour window burns quickly under video work.
• Standard: 4× Lite usage (10,000/7d, 3,000/5h) for about 3× the price. Most people should land here.
• Pro: for several simultaneous agents or long sessions. It does not remove the windows and does not change the interactive-only terms.

Buying rule: start one tier below your forecast, watch real consumption on the console’s My Subscriptions page for a week, and upgrade if you keep pausing — upgrades are prorated and the limits rise immediately. Downgrades are not supported; to move down, wait for expiry and repurchase.

Limits to understand before subscribing:

• Interactive use only. No automated scripts, application backends, batch processing, or background scheduled tasks; violations can suspend the subscription or revoke the key.
• Personal use only. One subscription per account, no sharing. You may hold an Individual and a Team Edition simultaneously.
• Region and data. The only region is Singapore and the only deployment mode is Global; prompts and outputs cross borders.
• No cancellation path documented. The docs state Individual “does not currently support cancellation” — the practical exit is to let it expire (disable auto-renewal). Refund terms were not documented on the pages reviewed; check your own order.
• Key changes after re-purchase. Expiry + repurchase rotates the API key; every configured tool must be updated.

Five-step Hermes quick setup

Assumes a macOS or Linux terminal (Windows via WSL2). This follows QwenCloud’s documented Token Plan endpoint and transport, using Hermes’ safer named-provider pattern for the key.

1. Install Hermes (skip if installed): curl -fsSL https://hermes-agent.nousresearch.com/install.sh | bash, then reload your shell (source ~/.bashrc or source ~/.zshrc). Verify with hermes --version.
2. Get your Token Plan credentials — generate the Token Plan API key on the console (it is shown in full only once — save it immediately). Note the base URL: OpenAI-compatible https://token-plan.ap-southeast-1.maas.aliyuncs.com/compatible-mode/v1, or Anthropic-compatible https://token-plan.ap-southeast-1.maas.aliyuncs.com/apps/anthropic.
3. Add a named provider to ~/.hermes/config.yaml (open it in an editor; add or replace only these entries):

   providers:
     qwen-token-plan:
       api: https://token-plan.ap-southeast-1.maas.aliyuncs.com/apps/anthropic
       key_env: QWENCLOUD_TOKEN_PLAN_API_KEY
       transport: anthropic_messages
       default_model: qwen3.8-max-preview
   
   model:
     provider: qwen-token-plan
     default: qwen3.8-max-preview

   Then add exactly one line to ~/.hermes/.env with an editor (no shell redirect): QWENCLOUD_TOKEN_PLAN_API_KEY=your-token-plan-api-key. The key never goes in config.yaml.
4. Verify the credential files without printing the key: ls -l ~/.hermes/.env should show -rw------- (else run chmod 600 ~/.hermes/.env), and config.yaml should contain no api_key: line.
5. Smoke test: hermes chat --provider qwen-token-plan -q "Reply with exactly: QWENCLOUD_OK" should reply with QWENCLOUD_OK.

Model switching: hermes chat --provider qwen-token-plan -m qwen3.7-max for a one-off run; /model inside a session; hermes model is the full setup wizard. On the Token Plan, stick to the exact model IDs in the supported list — the text models are qwen3.8-max-preview, qwen3.7-max, qwen3.7-plus, and qwen3.6-flash.

How the quotas actually work

1. A rolling 5-hour window. A 5-hour timer starts at your first invocation. At your tier’s cap (700 / 3,000 / 12,000 Credits) the service pauses until the full 5 hours elapse and the quota resets.
2. A rolling 7-day window. The same mechanics on a 7-day timer (2,500 / 10,000 / 40,000) — it is not a fixed calendar date.
3. Dual-window deduction. Every call counts against both windows simultaneously; either cap pauses the service, even if the other window has headroom. Unused Credits never carry over.

The pauses are the practical reality inside an agent: a long Hermes session can exhaust the 5-hour window mid-task, and a burst of video or image generation can exhaust the 7-day window in an afternoon. Video is especially sneaky — asynchronous tasks settle their Credits after completion, so several queued clips can land as one large deduction. When a window is exhausted, API calls are blocked — you are not silently billed pay-as-you-go. Your options: wait for the window to elapse, upgrade (prorated, limits raised immediately), or buy a Credit Pack (not subject to either window).

Models, media, and exclusions

The supported-models list is an exact-string allowlist — QwenCloud’s docs say to match model IDs character-for-character and never infer version compatibility. The text models are qwen3.8-max-preview, qwen3.7-max, qwen3.7-plus, and qwen3.6-flash; the list also includes qwen-audio-3.0-tts-plus (speech synthesis), glm-5.2 and deepseek-v4-pro / deepseek-v4-flash-0731 (text), and the Wan (wan2.7-image, wan2.7-image-pro) and HappyHorse (happyhorse-1.1-i2v / t2v / r2v) generation models. There are no qwen3-coder-* models on the Token Plan — an off-list ID returns “model not found or not supported.”

Two scope qualifications:

• Image and video generation models use a separate multimodal API that must be integrated through a tool’s Skill or extension mechanism — they are not reachable through the chat configuration in this guide.
• Native Harness tool calling is Responses-API-specific (documented for the qwen3.7 and qwen3.8 series only). The Anthropic-compatible transport used here is not documented to enable it — treat Harness tools as a separate integration.

qwen3.8-max-preview is a limited-time preview model: Credits consumption as low as 10% of the standard rate (a 10× usage boost), with an extra 80% off between 22:00 and 08:00 (UTC+8); it may be taken offline or replaced when the preview ends. Thinking is always enabled, temperature defaults to 0.6 (lower values are raised), and reasoning_effort accepts xhigh / medium / low (default xhigh).

Credit Packs ($15/pack/month, 20,000 Credits, up to 5, requires an active subscription) are valid for one month; unused pack Credits expire and are non-refundable.

Token Plan vs Coding Plan vs pay-as-you-go

Three QwenCloud billing surfaces share a family resemblance. Their API keys and base URLs are completely separate and must be used together:

In Hermes terms: the alibaba provider (DASHSCOPE_API_KEY) is the pay-as-you-go route; Token Plan is the named provider qwen-token-plan configured above; Coding Plan is a separate SKU on its own endpoint. QwenCloud’s FAQ confirms the pairing rule: a Token Plan key against the standard dashscope-intl base URL returns 401 Incorrect API key provided.

Command map

Check usage on the console’s Token Plan > My Subscriptions page (Token Plan Individual billing page): it shows the Credits quota and consumption of the current subscription, including both windows, and per-request deductions appear in the usage details there.

Troubleshooting

• 401 InvalidApiKey: Invalid API-key provided — wrong key type, expired subscription, or partial copy. Confirm the Token Plan Individual key is in ~/.hermes/.env as QWENCLOUD_TOKEN_PLAN_API_KEY.
• 401 Incorrect API key provided — Token Plan key paired with the standard dashscope-intl base URL. The provider’s api must be the token-plan base URL.
• 401 invalid access token or token expired — a Coding Plan or other billing mode’s base URL is in use; switch to the Token Plan base URL.
• 404 model '...' not found or not supported — the model ID is misspelled or not in the Token Plan allowlist; match the exact IDs, no inferred variants.
• 429 Requests rate limit exceeded — too many requests in a short window; wait about a minute and retry at lower frequency.
• 429 Allocated quota exceeded — a 5-hour or 7-day window is exhausted; wait, upgrade, or buy a Credit Pack. You are not billed pay-as-you-go when blocked.
• Mid-session pause during an agent run — this is the window cap, not a network fault. Check My Subscriptions; if it recurs, the plan shape is wrong for the workload.
• Peak-hour slowdowns — queuing is possible; QwenCloud suggests upgrading or Team Edition.
• Key stops working after a re-purchase — expiry rotates the API key; regenerate and reconfigure every tool.

Security and billing

Key discipline. The Token Plan API key is shown in full once. Store it in ~/.hermes/.env as QWENCLOUD_TOKEN_PLAN_API_KEY (mode 600), never inside config.yaml, and reset only if it leaks — resetting invalidates the key everywhere at once. Never paste it into chat, logs, or shell history. This named-provider + key_env pattern is Hermes’ documented security configuration combined with QwenCloud’s documented endpoint and transport — documentation-verified only; no authenticated Qwen key test was performed during drafting.

Billing. Prices are limited-time ($6/$18/$68; original $8/$25/$80) with monthly, quarterly, or yearly periods at checkout. Upgrades are prorated with immediate limit raises; downgrades require expiry and repurchase; auto-renewal can be disabled in the console; renewal extends the period only. The docs state Individual does not currently support cancellation, and refund terms were not documented on the pages reviewed — confirm against your own order at checkout. Credit Pack Credits expire after one month and are non-refundable.

Terms. Interactive use only — no automated scripts, application backends, batch processing, or background scheduled tasks. If you run Hermes on a schedule (cron-style jobs), that use conflicts with the Token Plan terms; scheduled work requires evaluating a billing surface that permits it, and the pay-as-you-go API must be assessed separately. Service region is Singapore with Global deployment, so prompts and outputs cross borders. For teams, QwenCloud offers Token Plan Team Edition.