GPT-4o mini Search Preview is a specialized model for web search in Chat Completions. It is trained to understand and execute web search queries.
GPT-4o-mini Search Preview is a variant of OpenAI's GPT-4o-mini model that includes built-in web search capabilities. It is designed to provide answers that incorporate real-time information from the…
The model combines general language understanding and generation with live web search. It can answer questions about recent events, pull data from specific websites, and synthesize information from multiple online sources. It retains typical GPT-4o-mini capabilities: text summarization, translation, code generation, logical reasoning, and creative writing. The search preview adds the ability to access current news, stock prices, weather, product details, and other time-sensitive data. However, because the model is tuned for speed and efficiency, its reasoning depth may be lower than that of the full GPT-4o model.
The model excels in scenarios where answers depend on changing or recent information. Examples include live customer service agents that need to check order status or return policies, content tools that generate summaries of trending topics, and research assistants that compile facts from multiple online sources. It is also useful for verifying claims against current data, such as checking a company's latest announcement or a sports score. Because of its relatively low cost and fast response times, it is suitable for applications with high request volumes where web grounding is a key feature.
If your application does not require web search or up-to-date information, consider using the base GPT-4o-mini without search preview. The base model is even less expensive (same per-token pricing, but without the overhead of search retrieval) and may be faster for purely conversational tasks. For very simple queries that don't need broad world knowledge, smaller models like GPT-4o-mini (base) or even older models like GPT-3.5 Turbo may suffice. Evaluate whether the search preview feature justifies the cost for your specific use case.
The search preview is automatically triggered by the model when it deems the query needs external information. You do not need to configure search APIs or pass in search results manually. However, you can influence the model's behavior by prompting it to use web search explicitly, or by specifying sources. The model respects standard safety and content filters applied by OpenAI. Note that web search may introduce latency compared to the base model, since the model waits for search results before generating the final response.
Specific benchmark scores for GPT-4o-mini Search Preview have not been publicly released as a separate variant. However, based on the base GPT-4o-mini, this model is expected to perform well on standard NLP benchmarks like MMLU, HellaSwag, and GSM8K, but likely with lower accuracy than the full GPT-4o. The search preview feature may improve performance on tasks requiring current knowledge (such as trivia about recent events) but does not change the underlying model's reasoning capabilities. For factual accuracy, the model relies on the quality and timeliness of the web sources it retrieves.
GPT-4o-mini Search Preview is designed for low latency compared to GPT-4o. The base GPT-4o-mini is known for fast inference, and the search preview adds extra time for web retrieval. Total response time depends on factors such as network latency, the number of search results retrieved, and the length of the context. In typical usage, responses are generated within a few seconds. For applications requiring absolute minimum latency, the base GPT-4o-mini (without search) would be faster. OrcaRouter's API supports streaming to begin displaying tokens as they are generated.
Strengths: The model provides up-to-date information without separate search integration, has a large context window (128k tokens), and is cost-effective for high-volume use. Limitations: It accepts text only; it may sometimes retrieve incomplete or irrelevant web content; its reasoning is less deep than GPT-4o; and the search preview can increase latency. Additionally, the model may not always cite sources explicitly, so users should verify critical information. It is not designed for tasks like multimodal understanding, advanced math, or complex chain-of-thought reasoning where GPT-4o performs better.
OrcaRouter bills GPT-4o-mini Search Preview at the provider rate with zero markup: $0.15 per million input tokens and $0.60 per million output tokens. This means you pay exactly what OpenAI charges, without any additional fees. Pricing is per token, with context tokens counted as input and generated tokens as output. The model uses the standard OpenAI tokenization scheme. There are no extra charges for the search preview feature itself; the cost is the same as the base GPT-4o-mini, despite the additional web retrieval.
Compared to GPT-4o (which costs $2.50 per million input tokens and $10 per million output tokens), GPT-4o-mini Search Preview is significantly cheaper—about 16 times less for input and 16.6 times less for output. This makes it a strong candidate for applications where cost per query matters, especially when you need web search. However, the base GPT-4o-mini (without search preview) costs the same per token but lacks search. If search is not needed, you could save the slight extra latency, but the per-token cost is identical.
OrcaRouter may support caching of search results or model responses, depending on the configuration. If enabled, repeated queries for the same information could be served from cache, reducing both latency and token usage. However, caching policies vary. For the most current information, the model typically performs a fresh web search on each request. To minimize costs, consider using the base GPT-4o-mini if your queries do not require web data. Always monitor your token usage through OrcaRouter's dashboard to understand your spending.
OrcaRouter passes through OpenAI's exact pricing without adding any margin. Billing is based on token counts as reported by the provider. There are no setup fees, monthly minimums, or extra charges for the API gateway. You only pay for the tokens you use at the rates specified. For GPT-4o-mini Search Preview, that means $0.15 per million input tokens and $0.60 per million output tokens. Ensure your account is set up with OrcaRouter to access the model at these rates.
To use the model, send a POST request to https://api.orcarouter.ai/v1/chat/completions with the model field set to "openai/gpt-4o-mini-search-preview". This endpoint is OpenAI-compatible, so you can use any OpenAI SDK or client by changing the base URL and API key. Example curl: curl https://api.orcarouter.ai/v1/chat/completions \ -H "Authorization: Bearer YOUR_ORCAROUTER_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "model": "openai/gpt-4o-mini-search-preview", "messages": [{"role": "user", "content": "What is the current weather in London?"}] }' The search preview will be triggered automatically.
The API supports all standard OpenAI chat completion parameters: temperature (0-2, default 1), top_p, n, stream, stop, max_tokens (up to 16384), presence_penalty, frequency_penalty, and logit_bias. The model respects these parameters. For search preview, there are no additional parameters needed; the model decides when to search. However, you can prompt it to explicitly request a web search (e.g., "Search the web for latest news about...") to encourage retrieval. Note that the model may still choose not to search if it deems internal knowledge sufficient.
If you already use OpenAI's API, migrating to OrcaRouter is straightforward: change the base URL from https://api.openai.com/v1 to https://api.orcarouter.ai/v1, and replace your API key with one from OrcaRouter. The model ID becomes "openai/gpt-4o-mini-search-preview". All other request bodies and response formats remain identical. You can use any OpenAI client library (Python, Node.js, etc.) without modification. OrcaRouter also supports streaming, function calling, and other advanced features compatible with OpenAI's API.
To stream responses, set "stream": true in your request. OrcaRouter returns data as server-sent events, identical to OpenAI's streaming format. Each event contains a delta of the response. Streaming is particularly useful for user-facing applications to show tokens as they are generated. For search preview queries, the model may first indicate that it is searching before outputting the final answer. The stream will include those interim tokens. Ensure your client can handle partial updates gracefully.
GPT-4o-mini Search Preview is a smaller, faster, and cheaper variant of GPT-4o with added web search capabilities. While GPT-4o has superior reasoning, larger knowledge base (up to its training cutoff), and multimodal support (images, audio), this model is text-only and relies on web search to overcome its limited training knowledge. For tasks requiring deep analysis or complex math, GPT-4o is better. For cost-sensitive applications needing fresh information, GPT-4o-mini Search Preview is a strong alternative. Pricing is about 16 times cheaper.
The base GPT-4o-mini does not include web search; it relies solely on its training data, which has a cutoff date in late 2023. GPT-4o-mini Search Preview adds the ability to retrieve current information from the internet. Both models have the same context window (128k tokens), max output (16384), and per-token pricing. The search preview variant may have slightly higher latency due to web retrieval. If your application does not need live data, the base model is functionally identical and may be faster. Both models are accessed via OrcaRouter with different model IDs.
Other search-enabled models include Gemini with Google Search grounding, or models using plugin-based web retrieval. GPT-4o-mini Search Preview offers native integration without external plugins. It is generally cheaper than larger search-enabled models (like GPT-4 with browsing). However, it may be less comprehensive in search results compared to dedicated search APIs. For high-accuracy web knowledge, consider supplementing with a factual search API and feeding results to a base model. This model is best for simple to moderately complex queries that benefit from current web data.
Choose this model when you need: (1) low cost per query, (2) automatic web search without custom integration, (3) fast response times, and (4) a large context window. Choose a full GPT-4o variant if you need multimodal inputs or deeper reasoning. Choose the base GPT-4o-mini if you don't need web search and want slightly faster responses. Choose a dedicated search API plus a small model if you need fine-grained control over search sources and results. This model is best as a drop-in replacement for GPT-3.5 or GPT-4o-mini in applications that would benefit from live information.
OpenAI-compatible — keep the SDK you already use
https://api.orcarouter.ai/v1from openai import OpenAI
client = OpenAI(
base_url="https://api.orcarouter.ai/v1",
api_key="$ORCAROUTER_API_KEY",
)
response = client.chat.completions.create(
model="openai/gpt-4o-mini-search-preview",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)max_tokensresponse_formatstreamstructured_outputsweb_search_options| Input / 1M tokens | $0.150 |
| Output / 1M tokens | $0.600 |
| Currency | USD |
Estimate based on list price
Estimate only — actual token counts depend on the provider's tokenizer.
GET /api/public/models/openai/gpt-4o-mini-search-previewOpen @misc{orcarouter_gpt_4o_mini_search_preview,
title = {GPT-4o-mini Search Preview API},
author = {OpenAI},
year = {2025},
howpublished = {OrcaRouter},
url = {https://www.orcarouter.ai/models/openai/gpt-4o-mini-search-preview}
}OpenAI. (2025). GPT-4o-mini Search Preview API. OrcaRouter. https://www.orcarouter.ai/models/openai/gpt-4o-mini-search-preview