Pricing · USD
Tools & additional fees
See when tools add to model usage and how each charge is counted.
Tool call prices
These are configured base rates. Model-specific rates and your account’s billing adjustments may change the final charge.
| Tool or service | Base price · USD | When a charge applies |
|---|---|---|
Standalone search endpoint/v1/alpha/searchA direct API request. Uses the web search preview rate and its model-specific overrides. | Default rate $10 per 1,000 requests $0.01 per request Model-specific rates Model ID starts with gpt-4.1-mini $25 per 1,000 requests $0.025 per request Model ID starts with gpt-4o-mini $25 per 1,000 requests $0.025 per request Model ID starts with gpt-4.1 $25 per 1,000 requests $0.025 per request Model ID starts with gpt-4o $25 per 1,000 requests $0.025 per request | Each successful request counts as one tool call. A model name ending in search-preview adds one more fixed search charge. Any configured model charge is separate. |
Built-in web searchweb_searchWeb search performed inside a model response. | $10 per 1,000 calls $0.01 per call | Counted from the search tool calls returned by the model. One model request can contain multiple searches. |
Web search previewweb_search_previewResponses uses this rate when web_search_preview is selected. It also takes priority if both search tools are declared, or if the search tool declaration is not recognized. | Default rate $10 per 1,000 calls $0.01 per call Model-specific rates Model ID starts with gpt-4.1-mini $25 per 1,000 calls $0.025 per call Model ID starts with gpt-4o-mini $25 per 1,000 calls $0.025 per call Model ID starts with gpt-4.1 $25 per 1,000 calls $0.025 per call Model ID starts with gpt-4o $25 per 1,000 calls $0.025 per call | Responses counts returned preview search calls. Outside Responses, models ending in search-preview add one fixed search charge per request. Rates may differ by model. |
File searchfile_searchThe file search tool in a Responses request. | $2.5 per 1,000 calls $0.0025 per call | Counted from returned file search tool calls. This line covers retrieval calls only. |
Google search groundingGoogle Search · GeminiSearch grounding within a Gemini response. | $14 per 1,000 requests $0.014 per request | One charge for a request that returns search information, regardless of the number of search queries. |
Responses image generationimage_generation · ResponsesImage generation used as a tool in a Responses request. | $150 per 1,000 completed image results $0.15 per completed image result | Counted from deduplicated final image results. Explicitly failed, incomplete, or canceled responses incur no image tool charge. Model usage is charged separately. |
A model ID uses its longest matching prefix rate. If no prefix matches, the default rate applies. A zero rate means this tool adds no surcharge; model usage can still be billed.
Custom function calls
Custom functions and tool_use calls have no default tool surcharge. An operator can configure a charge for a tool name; those charges are separate from this built-in tool price list. Model input and output usage still applies.
How charges add up
- Model usage and tool charges are added together. Tool charges can also apply when the model uses a fixed per-call price or expression-based pricing.
- Tool base rates are adjusted by the applicable account or model-group multiplier.
- Eligible promotions are applied after group adjustments. The final charge depends on the billing settings for your request.
- Failed requests and canceled tasks follow the rules of the service used. They are not universally free; a completed tool operation or a qualifying special charge may still be billed.
Tool charge examples
At the default rate, three completed web search calls add $0.03 before group adjustments and promotions. Model usage is charged separately.
At the default rate, two completed image generation results add $0.3 before group adjustments and promotions. Model usage is charged separately.
Special safety charge
This charge is enabled. It applies only when an upstream response matches a specific safety-error rule, including some failed requests. It does not apply to every refusal or error.
$0.05 base price per matching request
The applicable group multiplier affects this charge. Promotions do not apply.
Model and media usage
Token input and output, cached input, cache writes, audio usage, and media generation have model-specific pricing. Check the relevant model’s rates separately.