Overview
Bifrost computes request costs using a built-in pricing catalog that is automatically synced from a remote datasheet. Custom Pricing lets you override those catalog prices at runtime without redeploying, applying your own rates for any model across any combination of provider, key, and virtual key scopes.
Key capabilities:
- Scoped overrides — apply prices globally or narrow them to a specific provider, provider key, or virtual key
- Pattern matching — target an exact model name or a wildcard prefix (e.g.
gpt-4*)
- Request type filtering — restrict an override to one or more specific operations (chat, embeddings, image generation, etc.); at least one request type is required
- Hierarchical resolution — the most-specific matching override always wins; broader scopes act as fallbacks
Pricing data source
Before configuring overrides, Bifrost needs a pricing catalog to work from. By default it ships with built-in prices and syncs them every 24 hours. You can point it at a custom pricing URL if you maintain your own datasheet.
- Navigate to Models in the sidebar
- Click the Pricing Settings tab
- Enter your pricing datasheet URL in the Pricing Datasheet URL field
- Set the Pricing Sync Interval (in hours)
- Click Save
{
"framework": {
"pricing": {
"pricing_url": "https://your-host/pricing.json",
"pricing_sync_interval": 86400
}
}
}
| Field | Type | Required | Default | Description |
|---|
pricing_url | string (URI) | No | built-in | URL of the pricing datasheet to sync from |
pricing_sync_interval | integer | No | 86400 | Sync interval in seconds. Minimum 3600 (1 hour) |
Scope hierarchy
Every override is assigned a scope kind that determines which requests it applies to. When Bifrost resolves pricing for a request, it evaluates all matching overrides and selects the one with the most specific scope. More specific scopes always win over broader ones.
virtual_key_provider_key (most specific)
virtual_key_provider
virtual_key
provider_key
provider
global (least specific / catch-all)
| Scope kind | Required | Description |
|---|
global | — | Applies to every request regardless of provider, key, or virtual key |
provider | provider_id | Applies to all keys under a specific provider |
provider_key | provider_key_id | Applies to a specific provider API key only |
virtual_key | virtual_key_id | Applies to all requests made under a virtual key |
virtual_key_provider | virtual_key_id + provider_id | Applies when a virtual key routes to a specific provider |
virtual_key_provider_key | virtual_key_id + provider_key_id | Most specific: virtual key + exact provider API key |
Scope identifiers are exclusive to their scope kind — you cannot mix them. For example, virtual_key_provider requires virtual_key_id and provider_id and must not include provider_key_id.
Pattern matching
The pattern field controls which model names the override applies to. The match_type field controls how the pattern is interpreted.
| Match type | Behavior | Example |
|---|
exact | Matches only the exact model name | gpt-4o matches only gpt-4o |
wildcard | Prefix match — pattern must end with * | gpt-4* matches gpt-4o, gpt-4-turbo, gpt-4o-mini |
For wildcard patterns, append a * at the end of the prefix. For example, claude-3* will match all Claude 3 variants.
Request type filtering
request_types is required and must contain at least one value. Only request types that have pricing support are accepted. Stream variants are treated identically to their base type — specifying chat_completion covers both streaming and non-streaming chat requests.
| Type | Description |
|---|
chat_completion | Chat requests (streaming included) |
text_completion | Legacy text completions (streaming included) |
responses | Responses API requests (streaming included) |
embedding | Embedding generation |
rerank | Reranking |
speech | Text-to-speech (streaming included) |
transcription | Speech-to-text (streaming included) |
image_generation | Image generation (streaming included) |
image_variation | Image variation |
image_edit | Image editing (streaming included) |
video_generation | Video generation |
video_remix | Video remixing |
Creating an override
-
Navigate to Models → Pricing Overrides in the sidebar
-
Click Create Override
-
Fill in the form:
- Name — a human-readable label
- Scope — select the scope kind and provide the matching IDs
- Pattern — enter the model name or wildcard prefix
- Match type — choose Exact or Wildcard
- Request types — select one or more request types (required)
- Pricing fields — enter the price values you want to override (only non-zero fields are applied)
-
Click Save
curl -X POST http://localhost:8080/api/governance/pricing-overrides \
-H "Content-Type: application/json" \
-d '{
"name": "GPT-4o reduced input cost",
"scope_kind": "global",
"match_type": "exact",
"pattern": "gpt-4o",
"request_types": ["chat_completion"],
"patch": {
"input_cost_per_token": 0.0000025,
"output_cost_per_token": 0.000010
}
}'
{
"message": "Pricing override created successfully",
"pricing_override": {
"id": "550e8400-e29b-41d4-a716-446655440000",
"name": "GPT-4o reduced input cost",
"scope_kind": "global",
"match_type": "exact",
"pattern": "gpt-4o",
"request_types": ["chat_completion"],
"pricing_patch": "{\"input_cost_per_token\":0.0000025,\"output_cost_per_token\":0.00001}",
"created_at": "2026-03-20T10:00:00Z",
"updated_at": "2026-03-20T10:00:00Z"
}
}
curl -X PATCH http://localhost:8080/api/governance/pricing-overrides/{id} \
-H "Content-Type: application/json" \
-d '{
"patch": {
"input_cost_per_token": 0.000002
}
}'
curl -X DELETE http://localhost:8080/api/governance/pricing-overrides/{id}
# All overrides
curl http://localhost:8080/api/governance/pricing-overrides
# Filter by scope
curl "http://localhost:8080/api/governance/pricing-overrides?scope_kind=virtual_key&virtual_key_id=vk-abc123"
Pricing overrides are defined under governance.pricing_overrides. Each entry requires id, name, scope_kind, match_type, pattern, and request_types. The pricing_patch is a JSON-encoded string containing only the fields you want to override.{
"governance": {
"pricing_overrides": [
{
"id": "550e8400-e29b-41d4-a716-446655440000",
"name": "Global GPT-4o rate",
"scope_kind": "global",
"match_type": "exact",
"pattern": "gpt-4o",
"request_types": ["chat_completion"],
"pricing_patch": "{\"input_cost_per_token\":0.0000025,\"output_cost_per_token\":0.00001}"
},
{
"id": "660e8400-e29b-41d4-a716-446655440001",
"name": "All Claude models for prod VK",
"scope_kind": "virtual_key",
"virtual_key_id": "vk-abc123",
"match_type": "wildcard",
"pattern": "claude-3*",
"request_types": ["chat_completion"],
"pricing_patch": "{\"input_cost_per_token\":0.000003,\"output_cost_per_token\":0.000015}"
}
]
}
}
| Field | Type | Required | Description |
|---|
id | string | Yes | Unique override ID (UUID recommended) |
name | string | Yes | Human-readable label |
scope_kind | string | Yes | One of: global, provider, provider_key, virtual_key, virtual_key_provider, virtual_key_provider_key |
virtual_key_id | string | Conditional | Required for virtual_key* scopes |
provider_id | string | Conditional | Required for provider and virtual_key_provider scopes |
provider_key_id | string | Conditional | Required for provider_key and virtual_key_provider_key scopes |
match_type | string | Yes | exact or wildcard |
pattern | string | Yes | Model name or wildcard prefix ending with * |
request_types | array | Yes | Request types this override applies to. At least one value required. |
pricing_patch | string | No | JSON-encoded pricing fields to override |
config_hash | string | No | Auto-managed. Do not set manually |
Pricing fields reference
Only fields with non-zero values are applied. All values are cost per unit in USD.
Token costs
| Field | Description |
|---|
input_cost_per_token | Standard input token cost |
output_cost_per_token | Standard output token cost |
input_cost_per_token_batches | Input token cost for batch requests |
output_cost_per_token_batches | Output token cost for batch requests |
input_cost_per_token_priority | Input token cost for priority requests |
output_cost_per_token_priority | Output token cost for priority requests |
input_cost_per_character | Input cost per character (character-billed models) |
Token tier costs
| Field | Description |
|---|
input_cost_per_token_above_128k_tokens | Input cost above 128k context |
output_cost_per_token_above_128k_tokens | Output cost above 128k context |
input_cost_per_token_above_200k_tokens | Input cost above 200k context |
output_cost_per_token_above_200k_tokens | Output cost above 200k context |
Cache costs
| Field | Description |
|---|
cache_creation_input_token_cost | Cost to write a token to the prompt cache |
cache_read_input_token_cost | Cost to read a cached input token |
cache_creation_input_token_cost_above_200k_tokens | Cache creation above 200k context |
cache_read_input_token_cost_above_200k_tokens | Cache read above 200k context |
cache_read_input_token_cost_priority | Priority cache read cost |
cache_read_input_image_token_cost | Cache read cost for image tokens |
cache_creation_input_audio_token_cost | Cache creation cost for audio tokens |
Image costs
| Field | Description |
|---|
input_cost_per_image | Cost per input image |
output_cost_per_image | Cost per generated image |
input_cost_per_pixel | Cost per input pixel |
output_cost_per_pixel | Cost per output pixel |
input_cost_per_image_token | Cost per image input token |
output_cost_per_image_token | Cost per image output token |
output_cost_per_image_low_quality | Generated image — low quality |
output_cost_per_image_medium_quality | Generated image — medium quality |
output_cost_per_image_high_quality | Generated image — high quality |
output_cost_per_image_auto_quality | Generated image — auto quality |
output_cost_per_image_above_512_and_512_pixels | Generated image > 512×512 |
output_cost_per_image_above_1024_and_1024_pixels | Generated image > 1024×1024 |
output_cost_per_image_above_2048_and_2048_pixels | Generated image > 2048×2048 |
output_cost_per_image_above_4096_and_4096_pixels | Generated image > 4096×4096 |
Audio and video costs
| Field | Description |
|---|
input_cost_per_audio_token | Cost per audio input token |
input_cost_per_audio_per_second | Cost per second of audio input |
input_cost_per_second | Cost per second of input (generic) |
input_cost_per_video_per_second | Cost per second of video input |
output_cost_per_audio_token | Cost per audio output token |
output_cost_per_second | Cost per second of audio output |
output_cost_per_video_per_second | Cost per second of video output |
input_cost_per_video_per_second_above_128k_tokens | Video input cost above 128k context |
input_cost_per_audio_per_second_above_128k_tokens | Audio input cost above 128k context |
Other costs
| Field | Description |
|---|
search_context_cost_per_query | Cost per web search context query |
code_interpreter_cost_per_session | Cost per code interpreter session |
Examples
Flat rate for all Anthropic models
Apply a single input/output rate to every Claude model globally:
{
"id": "anthropic-flat-rate",
"name": "Anthropic flat rate",
"scope_kind": "provider",
"provider_id": "anthropic",
"match_type": "wildcard",
"pattern": "claude*",
"request_types": ["chat_completion", "text_completion", "responses"],
"pricing_patch": "{\"input_cost_per_token\":0.000003,\"output_cost_per_token\":0.000015}"
}
Per-virtual-key negotiated rate
A specific virtual key has negotiated lower prices for GPT-4o:
{
"id": "vk-prod-gpt4o-rate",
"name": "Prod VK — GPT-4o negotiated rate",
"scope_kind": "virtual_key",
"virtual_key_id": "vk-abc123",
"match_type": "exact",
"pattern": "gpt-4o",
"request_types": ["chat_completion"],
"pricing_patch": "{\"input_cost_per_token\":0.000002,\"output_cost_per_token\":0.000008}"
}
Image generation override
Override costs for a specific image model at global scope:
{
"id": "dall-e-3-rate",
"name": "DALL-E 3 custom rate",
"scope_kind": "global",
"match_type": "exact",
"pattern": "dall-e-3",
"request_types": ["image_generation"],
"pricing_patch": "{\"output_cost_per_image_high_quality\":0.04,\"output_cost_per_image_medium_quality\":0.02}"
}
Global catch-all for a new model
Use a global override to add pricing for a model not yet in the built-in catalog:
{
"id": "my-new-model-rate",
"name": "my-new-model pricing",
"scope_kind": "global",
"match_type": "exact",
"pattern": "my-new-model-v1",
"request_types": ["chat_completion"],
"pricing_patch": "{\"input_cost_per_token\":0.000001,\"output_cost_per_token\":0.000005}"
}
Next steps
- Virtual Keys — Attach virtual-key-scoped overrides to virtual keys for per-customer pricing
- Budget and Limits — Understand how costs are tracked against budgets
- Model Catalog — Deep dive into how pricing resolution and cost calculation work internally
