Use the live catalog as the source of truth
Model names, versions, upstream availability and billing rules can change. The public marketing page explains how billing works; the live pricing page is where you should confirm the exact model ID and current rate immediately before deployment.
Two models from the same provider can have different input, output, cache and long-context rates. Always select the complete model ID.
What determines the cost of an API call
| Component | What it measures | Why it matters |
|---|---|---|
| Input tokens | Instructions, context and conversation history sent to the model | Long prompts and retrieved documents increase input cost |
| Output tokens | Tokens generated by the model | Output rates are often higher than input rates |
| Cache reads | Previously cached prompt content reused by a supported model | Useful for repeated system prompts and shared context |
| Context tiers | Pricing that changes beyond a context threshold | A request can move into a higher tier even when the model ID stays the same |
| Model capabilities | Reasoning, images, audio, video, tools or file inputs | Special modalities and capabilities may use separate billing rules |
Estimate production cost with your own traffic
- Choose two or three exact model IDs from the live catalog.
- Remove secrets and personal data from a representative request set.
- Run the same requests against each candidate model.
- Record prompt tokens, output tokens, cache usage, latency, quality and errors.
- Calculate cost per successful task, not only cost per million tokens.
A lower token rate does not automatically mean a lower production cost. A model that needs more retries, longer prompts or manual review can cost more per completed task.
Pricing and request format are separate decisions
Choose the model based on quality, latency and cost, then confirm that it supports the request format your application uses. API Models documents OpenAI-compatible Responses and Chat Completions, Anthropic Messages and Gemini generateContent.
Pricing questions
Where can I see the current price for a model?
Open the live Models and Pricing page. It is the source of truth for model IDs, availability, endpoint support and current billing.
Do input and output tokens cost the same?
Not necessarily. Many models use separate input, output and cache rates. Some models also apply context-length tiers or model-specific formulas.
Can I estimate cost before migration?
Yes. Replay a redacted evaluation set and compare quality, latency and total cost for the exact model IDs you plan to use.