Gen3 Admin Financial Controls Provider Rate Cards

Provider Rate Cards (Financial Controls)

Start Here

1. Open the vendor’s official pricing page from the table below.
2. Map list prices into Model Pricing fields on Financial Controls [route: /dashboard/billing?tab=models|Model Pricing tab].
3. Prefer Reset … to online defaults when OpenRouter or LiteLLM already lists the model; use manual entry for private endpoints, negotiated rates, or unsupported metering.
4. Document the source in each row’s Source / notes field.
5. Save model pricing and validate spend on tenant Observability → Billing.

Why this matters

Provider rate cards change frequently. GT AI OS does not substitute your commercial contract—it gives you a structured place to mirror what you pay (or what you charge back) per token, audio second, character, or image. Mapping vendor units correctly prevents systematic under- or over-charging in infrastructure credits and tenant billing analytics.

Details

Use this article with Model Pricing for workspace mechanics and Financial Controls for infrastructure balance and storage meters.

Major providers and official pricing docs

Provider	Typical GT AI OS provider type	Official pricing documentation
OpenAI	openai	OpenAI API pricing
Anthropic	anthropic	Anthropic pricing
Google / Gemini	google, vertex, gemini	Google AI Gemini API pricing
Azure OpenAI	azure	Azure OpenAI Service pricing
AWS Bedrock	bedrock	Amazon Bedrock pricing
Mistral	mistral	Mistral AI pricing
Groq	groq	Groq pricing
Local / Ollama	ollama, on-prem adapters	No public API list price—use $0.00 token rates unless you allocate internal cost
OpenRouter	openrouter	OpenRouter pricing (also exposed via https://openrouter.ai/api/v1/models)

Always confirm region, enterprise discount, and cached-token pricing on the vendor site before you lock GT AI OS numbers.

Map provider rate card → Financial Controls fields

Vendor publishes	GT AI OS field	Notes
$/1M input tokens	Input token price / 1M	Chat, embed, vision (image_analysis) capabilities
$/1M output tokens	Output token price / 1M	Chat and vision; often 0 for embeddings
$/minute or $/hour audio	Audio price / hour on transcription rows	Runtime meters audio_seconds; LiteLLM may supply per-second costs converted to hourly display
$/1M characters (TTS)	Input character price / 1M on speech_synthesis	Maps to input_characters pricing method
$/image	Image price / generated image on image_generation	Per-image image_count method
Model not in online catalogs	Manual input/output or unit price + Enabled	Status becomes Manual after save

CSV import uses the same columns: inputPricePerMillion, outputPricePerMillion, unitPrice, requestType, pricingMethod, priceSource, active. See Model Pricing for the full column list.

Example: OpenAI GPT-4o (chat)

Public list pricing (verify on OpenAI API pricing before you save):

Vendor list price	Financial Controls value
$5.00 / 1M input tokens	Input token price / 1M = 5.00
$15.00 / 1M output tokens	Output token price / 1M = 15.00

Set Source / notes to OpenAI API pricing – GPT-4o – YYYY-MM-DD. Request type chat, pricing method token I/O.

Example: Anthropic Claude Sonnet–class (chat)

Anthropic publishes separate input and output token rates (verify on Anthropic pricing):

Vendor list price (illustrative)	Financial Controls value
$3.00 / 1M input tokens	Input token price / 1M = 3.00
$15.00 / 1M output tokens	Output token price / 1M = 15.00

Use the exact model name/key from Models (claude-sonnet-4-…, etc.) so online reset and CSV import align.

Example: Embeddings

Vendor list price	Financial Controls value
$0.13 / 1M input tokens (text-embedding-3-large)	Input token price / 1M = 0.13 on embed capability
No output charge	Output token price / 1M = 0.00

Example: Speech and images

Capability	Vendor unit	Financial Controls
STT	Vendor $/minute → convert to $/hour if the UI labels hourly	Audio price / hour on transcription
TTS	Vendor $/1M characters	Input character price / 1M on speech_synthesis
Image gen	Vendor $/image	Image price / generated image on image_generation

When online reset marks a row Unsupported, the vendor doc still lists a price but GT AI OS cannot ingest that shape automatically—enter the unit field manually.

Reset to online defaults vs manual entry

Situation	Recommended approach
Model routed through OpenRouter with public catalog entry	Reset … to online defaults for fallback catalog floor rates; actual chat billing uses OpenRouter usage.cost at settlement
Direct OpenAI / Anthropic / Groq / Bedrock with LiteLLM coverage	Reset … to online defaults (LiteLLM API + raw JSON fallback)
Enterprise Azure deployment with private rates	Manual entry from your Azure rate card
Local Ollama / on-prem GPU	Manual $0.00 / $0.00 unless you allocate internal chargeback
Negotiated discount not reflected in public catalogs	Manual entry; note contract in Source / notes
Row Unsupported after reset	Manual unit or token prices

Online sources:

• https://openrouter.ai/api/v1/models
• https://api.litellm.ai/model_catalog
• https://raw.githubusercontent.com/BerriAI/litellm/main/model_prices_and_context_window.json

Manual rows are preserved during background model sync until you run an online reset on those rows.

Compound and multi-vendor models

• Groq Compound (groq/compound): price underlying models in Model Pricing or override in the compound section; see Model Pricing compound discussion.
• OpenRouter slugs (openai/gpt-4o): ensure provider type is OpenRouter so reset uses the OpenRouter catalog. Catalog sync imports the lowest listed prompt/completion rates from GET /api/v1/models (OpenRouter’s documented catalog floor across endpoints). That is a fallback for reservation estimates and settlement when usage.cost is absent; settled chat usage primarily tracks OpenRouter’s per-response usage.cost. See Model Pricing — OpenRouter.

Best practices

• Date-stamp Source / notes when vendors change list prices.
• Re-run online reset after LiteLLM or OpenRouter catalog updates, then diff remaining manual rows.
• Align Models model keys with vendor IDs before importing CSV.
• Use Storage pricing for GiB-month retention; do not mix storage into token rows.

GT Helper example questions

Ask GT Helper from Pricing Guides or the ? shelf:

• “Map Azure OpenAI gpt-4o mini list prices to Financial Controls input and output per million.”
• “Should I use online reset or manual pricing for Bedrock Claude on a private account?”
• “What price should I set for local Ollama llama3 in model pricing?”
• “How do I enter Groq Whisper STT pricing when the row is unsupported?”

Related pages

• Model Pricing
• Financial Controls
• Storage pricing
• Models
• Observability
• GT Helper