Model Pricing

Configure the per-token cost of each model so Hrida AI Studio can track spend, enforce budgets, and surface cost data in analytics.

Model pricing lets admins record the input and output token costs for every LLM connected to the instance. Once configured, Hrida AI Studio automatically calculates the USD cost of each message and persists it alongside the conversation — feeding the cost columns in Analytics, and powering per-user and per-group Usage Limits.

Admin only

All model pricing endpoints require an admin account. Regular users can view their own accumulated cost via the Usage Limits GET /me endpoint.

---

How it works

1. Admin sets input_cost_per_million and output_cost_per_million (in USD) for a model.
2. When a chat message is saved, the backend reads the token counts from the model response, looks up the pricing entry by model_id, and stores cost_usd on the message row.
3. The Analytics dashboard aggregates these costs. Usage Limits reads them to enforce quotas.

Cost calculation uses eight decimal places of precision (0.00000001) to accurately represent sub-cent amounts for cheap or short messages.

---

Default prices

When the instance starts for the first time, the model_pricing table is automatically seeded with pricing for well-known models including:

• OpenAI: gpt-4o, gpt-4o-mini, gpt-4, gpt-3.5-turbo, o-series reasoning models
• Anthropic: claude-3-5-sonnet-*, claude-3-opus-*, claude-haiku-*, claude-sonnet-4-6, etc.
• Google: Gemini models
• Common open-source models via Ollama pricing approximations

If the table already has entries when the instance starts, seeding is skipped — your customizations are preserved.

---

Managing pricing

Viewing current prices

Go to Admin Panel > Model Pricing to see the full table, or use the API:

GET /api/v1/model-pricing
Authorization: Bearer <admin-token>

Response:

[
  {
    "id": "uuid",
    "model_id": "gpt-4o",
    "provider": "openai",
    "input_cost_per_million": "2.50000000",
    "output_cost_per_million": "10.00000000",
    "currency": "USD",
    "is_active": true
  }
]

Setting or updating a price

Use an upsert — the same endpoint creates or updates:

PUT /api/v1/model-pricing/gpt-4o
Authorization: Bearer <admin-token>
Content-Type: application/json

{
  "provider": "openai",
  "input_cost_per_million": 2.50,
  "output_cost_per_million": 10.00,
  "currency": "USD",
  "is_active": true
}

The model_id is taken from the URL path and supports slashes (e.g., anthropic/claude-3-5-sonnet-20241022).

Disabling a price without deleting it

Set "is_active": false in the upsert payload. Inactive entries are ignored by the cost calculator — messages for that model will have cost_usd: null.

Deleting a price

DELETE /api/v1/model-pricing/{model_id}
Authorization: Bearer <admin-token>

---

API reference

Method	Path	Description
GET	/api/v1/model-pricing	List all pricing entries
PUT	/api/v1/model-pricing/{model_id}	Create or update pricing for a model
DELETE	/api/v1/model-pricing/{model_id}	Delete pricing for a model

---

Tips

• Model IDs must match exactly. The cost calculator does a direct lookup — if your model is registered as gpt-4o-2024-08-06 you need a pricing entry for that exact string, not gpt-4o.
• Currency is always USD in the current implementation. The currency field is stored for forward compatibility.
• Costs appear in analytics after the next analytics refresh. Already-saved messages with cost_usd: null are not back-filled.
• Combine with Usage Limits to turn cost tracking into enforcement — see Usage Limits.