CRITICAL — Before starting, you MUST use the Read tool to read ../arkcli-shared/SKILL.md, which contains the authentication gate, configuration troubleshooting, and shared security rules.
Applicable scenarios
- Query the settlement unit price of foundation models (various ChargeItems such as input / output / batch inference / context caching)
- View free quota (
InferenceFreeUsage、ResourcePackItems) and arrears status (IsOverdue) - View sub-service activation status (
SubServices:base/context-cache/fast-infer) - Filter the price catalog by modality
- Query AgentPlan / CodingPlan package subscription prices(personal edition + enterprise edition, 12 tiers in total)
Business positioning
- "How much does it cost / price inquiry" goes to pricing; "buy/renew" goes to arkcli-plans parameters and protocol flow; "what has already been purchased" uses
plans get, and the subscription holdings list must not be used as a price catalog. - "How much has already been spent" goes to arkcli-billing; "how much quota is left" goes to Usage.
usage balanceThe free quota, media asset library, and package balance are not cash balances or vouchers. - A quote must simultaneously state the currency, billing unit, applicable model/training type, tier, and current account discount; a missing price does not mean free, and a single item failure must not cause the entire catalog to be reported as empty. Do not use historical prices or multiply the original price by the discount again to replace real-time
Price。
- The two types of pricing are managed separately:Models are billed by usage(pricing models) and package subscription prices(pricing plans) are two completely independent commands. For the decision, see the "model vs package" section below
- Covers only foundation models (foundation models) catalog-style pricing. However, the ChargeItems of a base model are the full scope— they include not only the foundation model inference price, but also the training rate when using it for fine-tuning (
Finetune/LoraFinetune), as well as the inference price when calling the custom model produced by fine-tuning (FinetuneInference*/FinetuneI2I*, etc.). Therefore, querying a base model once retrieves all related rates. - Backend API does not return a separate entry for custom models (fine-tuned
cm-*) — the rates are inherently embedded in the base model's ChargeItems array. However, arkcli already provides reverse-lookup sugar:pricing models --model cm-xxxautomatically calls GetCustomModel to obtain the base name and then queries the price, and the top level of the response has an extraresolved_custom_modelfield (customization_type+foundation_model_name/foundation_model_version), and the Agent can select the corresponding ChargeItem.Type accordingly, without needing to manually perform a two-step process - Does not cover actual consumption at the Endpoint dimension → use arkcli-usage
- Does not include rate limits / token limits / context window: these are within the scope of
arkcli-models, go to../arkcli-models/SKILL.md. pricing only handles "money" - Audio catalog prices and invocation capabilities are separate：
pricing models --modality Audiois a valid query; it can be further filtered using a confirmed foundation model name. Only when the correspondingChargeItemsis returned should its price, currency, unit, and applicable conditions be answered; if the catalog is empty or a pricing item is missing, state that the current price source does not cover it, and do not assume it is free, nor promise support for TTS/ASR invocation, deployment, or usage statistics based on it. - Price = final unit price including the current account's discount(the backend has already calculated it based on contract/campaign/package), no client-side discount calculation is needed
- OriginalPrice = publicly listed original price;
Price < OriginalPricewhich means the account has a discount
Command
| Command | Description |
|---|---|
pricing models |
List the base model billing unit price catalog (pay-as-you-go, charged by token) |
pricing plans |
List subscription package prices (AgentPlan / CodingPlan Personal Edition + Enterprise Edition) |
Model vs Package — which command to use
Is the user asking about a "per-token unit price" or a "monthly package price"? This is the key distinguishing point.
| User wording | Should use | Reason |
|---|---|---|
| "How much is DeepSeek-V4" / "How much is doubao-seed" / "GPT-X price" | pricing models |
Model name → charged by token |
| "How much is image-to-image per image" | pricing models |
Charged per call, in ChargeItems |
| "How much is audio recognition per minute" / "How much is TTS" / "voice model cost" | pricing models --modality Audio |
Check the real ChargeItems; quote by the returned unit, do not default to per-minute or per-token |
| "How much is fine-tuning" / "How much token cost to fine-tune a model" | pricing models |
The fine-tuning rate is embedded in the base model ChargeItems (see Finetune / LoraFinetune type) |
| "How much does my cm-xxx call cost" / "How much is this fine-tuned model" | pricing models --model cm-xxx |
The CLI automatically looks up the base + exposes resolved_custom_model.customization_type, and extracts the ChargeItem according to the table below |
| "How much is Agent Plan" / "How much is Agent Plan Personal Edition small" | pricing plans |
Package name → monthly subscription |
| "How much is Coding Plan" / "lite / pro package price" | pricing plans |
Same as above |
| "Agent Plan Enterprise Edition price" / "How much is the team edition" | pricing plans --edition enterprise |
Same as above |
Confusion fallback: the user says "how much is X" and X is a model name → models; X is a plan name → plans; if neither seems right → first pricing plans list the 12 plans to see if there is a hit, then go back to models.
--modality value mapping (the Agent side is responsible for translation)
The backend FoundationModelDomain field is5 coarse-grained enums, arkcli passes it through directly without translation. When the user says "video / image / text" in natural language, you (the Agent) need to translate first and then pass it:
| The user might say | Actual backend enum | Note |
|---|---|---|
| Text / text / LLM / language model | LLM |
|
| Image / picture / image / visual / vision | ComputerVision |
|
| Video / video | ComputerVision |
⚠️ The backend does not distinguish image/video/3D,the return includes both image and 3D models |
| 3D / hyper3d | ComputerVision |
Same as above |
| Audio / audio / voice / voice / TTS / ASR | Audio |
Only quote if there is a catalog price; do not extrapolate calling capability |
| Vector / embedding / embedding | Embedding |
|
| Router / router | Router |
Important: when the user says "how much is video", you should:
- Call
arkcli pricing models --modality ComputerVision(cannot directly--modality video) - Filter out video-type models in the response (see
DisplayNameorFoundationModelTag), or explicitly tell the user "the backend returns by broad visual category, including image / video / 3D"
ChargeItem.Type key enum reference (essential for response parsing)
Each base model's ChargeItems isfull-scope—it may mix rates for basic inference, fine-tuning training, fine-tuning inference, and various generation tasks. The Agent picks only the relevant 1-2 entries based on the user's question.
| User question semantics | Which Type to extract |
|---|---|
| How much is inference | InferencePrompt + InferenceCompletion |
| How much is batch inference | BatchInferencePrompt + BatchInferenceCompletion |
| How much is context caching | ContextSessionHit + ContextSessionStorage |
| How much is fine-tuning | Finetune(full) or LoraFinetune(LoRA) |
| How much is calling the fine-tuned model | FinetuneInferencePrompt + FinetuneInferenceCompletion |
| How much is image-to-image (base model) | T2ICompletion / I2ICompletion |
| How much is image-to-image (fine-tuned model) | FinetuneT2ICompletion / FinetuneI2ICompletion |
| Video generation | T2VCompletion / I2VCompletion / V2VCompletion(and 1080/4K variants) |
| 3D generation | To3DCompletion / FinetuneTo3DCompletion |
| Low-latency inference | FastInferencePrompt / FastInferenceCompletion |
For the complete Type enum, see references/arkcli-pricing-models.md the "ChargeItem.Type full enum reference" section.
CustomizationType → ChargeItem.Type mapping
pricing models --model cm-* The response top level of resolved_custom_model.customization_typewill include references/arkcli-pricing-models.md the "Custom model (cm-*) reverse lookup" section.
Quick decision
- User asks "how much is model X":
arkcli pricing models --model X, fromChargeItemsextractInferencePrompt/InferenceCompletion - User asks "how much is my cm-xxx":
arkcli pricing models --model cm-xxx, look at the responseresolved_custom_model.customization_type→ extract according to the mapping table aboveFinetune*that entry;resolved_custom_model.foundation_model_namealso tell the user what its base is - User asks "all LLM prices":
arkcli pricing models --modality LLM - User asks "how much are TTS / ASR / speech models": do not execute pricing; redirect to
arkcli models search <keyword>only explain discoverability and that cost queries are currently unsupported - User asks "how much is Agent Plan / Coding Plan":
arkcli pricing planslist all 12 tiers, or with--plan agent-plan-personal-smallquery a single one - User asks "what free quota do I have":
pricing models, look atInferenceFreeUsageandResourcePackItems[Type=FreeInference] - User asks "does my account have a discount": compare
PricevsOriginalPrice, the difference is the discount (both commands support it) - User asks "my actual consumption": this is not a pricing matter, redirect to arkcli-usage
- Enterprise plan has no permission:
pricing planswill fill theerrorfield of those 6 enterprise rows withAccessDenied: ..., while the other 6 personal rows are returned normally. This is expected behavior, not a bug
Common fallbacks
- Authentication error: redirect to
../arkcli-auth/SKILL.md - User is asking about consumption (usage that has occurred) rather than unit price: redirect to
../arkcli-usage/SKILL.md - The audio price source has no target model/pricing item: explain the query scope and the gap; if the user still needs a quote for that product, redirect to its official billing documentation for verification, and do not apply another model's price to it.
Reference
- arkcli-usage — actual token / request consumption (including historical ranges)
- arkcli-deploy — create Endpoint (automatically detect activation before deployment)
- arkcli-shared — authentication and global parameters