Model groups & pricing

Models on DimiLinks belong to different groups, and the selected group directly determines the final cost. The following constraints apply before a request can run:

Each model is available only in a fixed set of groups. For example, claude-opus-4-8 is available only in Claude groups; the default group cannot run Claude.
Your API key has its own set of available groups. That set must intersect the target model’s available groups for the request to succeed.
You cannot extrapolate the cost of the same prompt linearly across models or groups. Billing uses “official price × model multiplier × group multiplier,” and combinations may differ by dozens of times.

Do not hardcode multipliers. Groups and prices may change at any time. Treat the GET /api/pricing response as the runtime source of truth.

Authoritative endpoint

curl https://api-direct.dimilinks.com/api/pricing

This public endpoint requires no authentication and returns all live groups, public group descriptions, models, and multipliers:

{
  "success": true,
  "pricing_version": "a42d372ccf0b5dd13ecf71203521f9d2",
  "group_ratio": {
    "default": 1,
    "open ai 特价": 0.5,
    "claude 特价": 0.3,
    "claude-max": 0.8,
    "gpt-image-2": 1
  },
  "usable_group": {
    "default": "",
    "open ai 特价": "open ai 自有号池",
    "claude 特价": "claude 自有号池",
    "claude-max": "claude 满血渠道",
    "gpt-image-2": ""
  },
  "auto_groups": [],
  "supported_endpoint": {
    "openai":      { "path": "/v1/chat/completions",   "method": "POST" },
    "anthropic":   { "path": "/v1/messages",           "method": "POST" },
    "image_generation": { "path": "/v1/images/generations", "method": "POST" },
    "image_edits": { "path": "/v1/images/edits", "method": "POST" }
  },
  "data": [
    {
      "model_name": "gpt-5.5",
      "enable_groups": ["open ai 特价", "default"],
      "model_ratio": 2.5,
      "completion_ratio": 6,
      "cache_ratio": 0.1,
      "quota_type": 0,
      "model_price": 0,
      "supported_endpoint_types": ["openai"]
    },
    {
      "model_name": "claude-opus-4-8",
      "enable_groups": ["claude 特价", "claude-max"],
      "model_ratio": 2.5,
      "completion_ratio": 5,
      "cache_ratio": 0.1,
      "quota_type": 0,
      "model_price": 0,
      "supported_endpoint_types": ["anthropic", "openai"]
    },
    {
      "model_name": "gpt-image-2",
      "enable_groups": ["gpt-image-2", "default"],
      "model_ratio": 0,
      "completion_ratio": 0,
      "cache_ratio": null,
      "quota_type": 1,
      "model_price": 0.08,
      "resolution_options": [
        { "resolution": "1k", "model_price": 0.08, "max_n": 1 },
        { "resolution": "2k", "model_price": 0.15, "max_n": 4 },
        { "resolution": "4k", "model_price": 0.30, "max_n": 4 }
      ],
      "supported_endpoint_types": ["image_generation", "image_edits"]
    }
  ]
}

Field definitions

Field	Meaning
`group_ratio`	Multiplier map for all current groups; the sole authoritative source
`usable_group`	Public descriptions for groups, suitable for documentation or UI display
`auto_groups`	The server’s allowlist for automatic groups; only groups listed here may be resolved to a specific model through automatic selection
`data[].enable_groups`	Groups supported by the model; the request must match at least one
`data[].model_ratio`	Input-token multiplier relative to the official baseline
`data[].completion_ratio`	Output-token multiplier, applied as a multiple of `model_ratio`
`data[].cache_ratio`	Multiplier for cache-hit input; `null` means the model does not distinguish cached input
`data[].quota_type`	`0` for token-based billing; `1` for per-request billing
`data[].model_price`	Price per request when `quota_type=1`
`pricing_version`	Billing snapshot version, used as a client cache key; retrieve fresh data when it changes

How callers select a group

There are two ways to select a group in a request:

Omit group (recommended): The server calculates “API key groups ∩ model enable_groups,” then selects one using auto_groups and internal priority. This default workflow is sufficient for most callers.
Pass group explicitly: Add a value such as "group": "claude 特价" to the request body. The group must belong to both the API key’s available groups and the model’s enable_groups; otherwise, the API returns 403 model_not_allowed.

curl https://api-direct.dimilinks.com/v1/chat/completions \
  -H "Authorization: Bearer $DIMILINKS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "claude-opus-4-8",
    "messages": [{"role": "user", "content": "Hello"}],
    "group": "claude 特价"
  }'

Specify group only when one API key has multiple eligible target groups and your application must consistently use one of them.

Billing formulas

For token-billed models (quota_type = 0):

input_cost  = input_tokens         × model_ratio                    × group_ratio
output_cost = output_tokens        × model_ratio × completion_ratio × group_ratio
cached_cost = cached_input_tokens  × model_ratio × cache_ratio      × group_ratio   (enabled when cache_ratio is not null)
total_cost  = input_cost + output_cost + cached_cost

For per-request models (quota_type = 1):

total_cost = model_price × group_ratio × n

n is the number of images or videos actually produced by the task. For image endpoints, it is the n field in the request body.

Cost examples

claude-opus-4-8 in the claude 特价 group, with 1K input tokens and 500 output tokens:

model_ratio = 2.5, completion_ratio = 5, group_ratio = 0.3
input = 1000 × 2.5 × 0.3 = 750
output = 500 × 2.5 × 5 × 0.3 = 1875
total ≈ 2625 in internal system quota, converted to display currency using the account’s quota_per_unit

Comparing gpt-5.5 in the default and open ai 特价 groups:

default: group_ratio = 1; the full model_ratio 2.5 × completion_ratio 6 calculation applies.
open ai 特价: group_ratio = 0.5; the final cost is half the cost in the default group.

One asynchronous 2K image from gpt-image-2:

The 2K tier has model_price = 0.15; group_ratio is either gpt-image-2 = 1 or default = 1, so the request costs 0.15. Read the separate resolution_options values for the 1K and 4K tiers instead of applying the 2K price.

Video billing: two groups, two rules

Video capability is split into two groups with different billing. Confirm which group a model belongs to before calling it; see Video groups and billing rules.

The `seedance/h3` group: per-second billing (tiered_expr)

minimax-h3, seedance 2.0, and seedance 2.5 use per-second billing: cost = per-second price × seconds, with the resolution determining the per-second price (seedance/h3 group ratio = 1). In /api/pricing, these models carry billing_mode: "tiered_expr" and a billing_expr that declares the per-second price of each resolution tier:

Model	480p	720p	1080p
`minimax-h3`	¥0.16/s	¥0.24/s	¥0.70/s
`seedance 2.0`	¥0.60/s	¥1.00/s	¥1.80/s
`seedance 2.5`	¥0.70/s	¥1.30/s	unsupported

The `doubao seedance` group: per-token billing

doubao-seedance-2-0-260128, doubao-seedance-2-0-fast-260128, doubao-seedance-2-0-mini-260615, and doubao-seedance-2-5-260628 use per-token billing (quota_type = 0) and support 720p only. The doubao seedance group ratio is 0.7, i.e. 30% off the official baseline. Token usage depends on duration × resolution and follows the standard token formula.

Per-resolution image billing

Model	1k	2k	4k
`gpt-image-2`	¥0.08/image	¥0.15/image	¥0.30/image
`nano-banana-pro`	¥0.12/image	¥0.18/image	¥0.195/image

gpt-image-2.5-flare / gpt-image-2.5-sunburst are billed by token ratio (input 37.5×, output 2×); the actual cost depends on the image size.

Engineering recommendations

Retrieve GET /api/pricing at startup and build a model_name → { enable_groups, model_ratio, completion_ratio, cache_ratio, quota_type, model_price } index. Use pricing_version as the cache key.
When showing available groups for a model in the UI, read data[].enable_groups and the usable_group descriptions directly instead of maintaining a separate hardcoded map.
Cost estimates must account for all four values: model_ratio, completion_ratio, cache_ratio, and group_ratio. Omitting any of them produces an inaccurate estimate.
When you see 403 model_not_allowed, first check whether the API key’s available groups cover the target model’s enable_groups. As a temporary workaround, set group explicitly to a group the current key actually has.

Basics

Chat

Images / gpt-image-2

Images / more models

Videos / unified video API

Video / grok-imagine-video

IDE and CLI integrations

Authoritative endpoint

Field definitions

How callers select a group

Billing formulas

Cost examples

Video billing: two groups, two rules

The `seedance/h3` group: per-second billing (tiered_expr)

The `doubao seedance` group: per-token billing

Per-resolution image billing

Engineering recommendations

​Authoritative endpoint

​Field definitions

​How callers select a group

​Billing formulas

​Cost examples

​Video billing: two groups, two rules

​The seedance/h3 group: per-second billing (tiered_expr)

​The doubao seedance group: per-token billing

​Per-resolution image billing

​Engineering recommendations

Authoritative endpoint

Field definitions

How callers select a group

Billing formulas

Cost examples

Video billing: two groups, two rules

The `seedance/h3` group: per-second billing (tiered_expr)

The `doubao seedance` group: per-token billing

Per-resolution image billing

Engineering recommendations