Preview text-only GPT-4o model with 128K context, 16K output, accessed via OrcaRouter API at $2.50/$10 per M tokens.
This model is a preview release from OpenAI, limited to text-only inputs. It carries the identifier openai/gpt-4o-search-preview-2025-03-11 when accessed via OrcaRouter's API. The model provides a…
As a text-only GPT-4o preview, it excels at understanding and generating natural language across a wide range of domains. It can perform document summarization, question answering, creative writing, translation, and code generation. The large context window allows it to maintain coherence over very long passages, making it suitable for tasks like book analysis or processing extensive logs. It supports system, user, and assistant messages in a chat-like format. The model's zero-shot and few-shot reasoning capabilities are typical for GPT-4 class models, enabling it to handle complex instructions without extensive fine-tuning.
If your task does not require the full reasoning power of GPT-4o, consider using a smaller or cheaper model from OrcaRouter, such as openai/gpt-4o-mini or gpt-3.5-turbo. Examples of tasks better suited for cheaper models include simple classification, short-form content generation, or basic Q&A with limited context. The preview model's cost may be unjustified for high-volume, low-complexity requests. Additionally, if multimodal input is required, this model cannot handle images or audio; another model (e.g., standard GPT-4o or GPT-4V) would be necessary. Always evaluate the trade-off between quality and cost per token.
This model is identified as a 'search-preview' variant, but no specific tool-use or web-search capabilities are detailed in the publicly available specifications. It is text-only and operates as a standard language model. Users should not assume any built-in search or retrieval functionality. For augmented generation, you can implement retrieval-augmented generation (RAG) pipelines separately, feeding retrieved documents into the model's context window. OrcaRouter does not add any tool-use middleware; the model responds solely based on its training and the prompt provided. Future versions may change this behavior, but as of the preview, no tool integration is documented.
As a preview, this model may exhibit unexpected behavior or lower reliability compared to stable releases. It is text-only, so any multimodal request will be rejected. The maximum output of 16,384 tokens is substantial but may still truncate very long generations. The model's knowledge cutoff and training data are not specified; it may not be aware of very recent events. Production use is discouraged because the model could be updated or removed. Additionally, latency may be higher than stable models due to its preview nature. Always test thoroughly before deploying at scale.
No specific benchmark scores for this particular preview version have been provided. As a GPT-4o variant, it is expected to perform similarly to the standard GPT-4o on common language tasks, but the exact numbers are not disclosed. Users can run their own evaluations using public datasets. The large context window gives it an advantage on long-document benchmarks, but since this is a preview, scores may differ from the production model. For authoritative numbers, refer to OpenAI's official benchmark reports for GPT-4o. OrcaRouter does not modify the model's outputs or performance.
Latency and throughput figures for this preview model have not been published. As a preview, it may be slower than the stable GPT-4o due to experimental infrastructure. Response times depend on input length, output length, and concurrent demand. OrcaRouter does not impose additional latency beyond normal API routing. To estimate speed, developers can run sample requests with typical payloads. For production applications requiring consistent performance, consider using the stable gpt-4o model instead, unless the preview's specific features are needed.
GPT-4 class models are known for strong reasoning across various domains, including math, logic, and common-sense inference. This preview is expected to inherit those capabilities. The large context enables complex chain-of-thought reasoning over many steps without losing track of earlier parts. However, no specific reasoning benchmarks are provided for this exact version. Users should validate performance on their own domain-specific tasks. The model may also exhibit improved instruction following compared to earlier GPT versions, but again, independent verification is recommended.
Beyond being a preview, the model has no multimodal support. It cannot process images, audio, or video directly. Its knowledge cutoff date is unknown, so it may lack information about events after its training. The 16,384 token output limit, while generous, cannot produce extremely long texts in one generation; for longer outputs, iterative prompting or splitting is needed. The preview may also have higher error rates or refusal rates for certain sensitive topics. Performance on languages other than English may vary. Always test on representative data before deployment.
The model is priced at $2.50 per 1 million tokens for input and $10.00 per 1 million tokens for output. These are OpenAI's provider rates; OrcaRouter adds no markup. For example, processing a prompt of 10,000 tokens and generating 1,000 tokens would cost approximately $0.025 for input and $0.01 for output, totaling $0.035. Costs scale linearly with token usage. There are no additional fees for using OrcaRouter as the API gateway. Be sure to monitor usage to avoid unexpected charges. Compare with cheaper models if your task does not need the full capabilities.
OrcaRouter does not currently offer volume discounts or token caching for this model. All billing is pay-as-you-go based on actual token usage at the provider rate. Caching of prompts or responses is not performed by OrcaRouter; each request is processed independently. If you have very high throughput, contact OrcaRouter for potential custom arrangements. For cost optimization, consider using shorter prompts, reusing context efficiently, or batching calls. The model does not support reduced pricing for cached hits since there is no caching mechanism in place.
At $2.50/$10 per million tokens, this preview is priced similarly to the standard GPT-4o but may differ slightly. Cheaper alternatives include openai/gpt-4o-mini (approximately $0.15/$0.60 per million) or gpt-3.5-turbo (roughly $0.50/$1.50). For tasks that can tolerate lower quality, those models offer significant savings. The preview model's cost is on the higher end of the OrcaRouter catalog. If you are using it solely for testing, you may incur moderate costs depending on request volume. Always evaluate expected token usage and compare with the quality gain from using a larger model.
To use this model, set your API client's base URL to https://api.orcarouter.ai/v1, and specify the model ID as openai/gpt-4o-search-preview-2025-03-11. You must use an API key issued by OrcaRouter. The API is fully OpenAI-compatible, so existing code written for OpenAI's chat completions endpoint will work with minimal changes. Example call using curl: curl https://api.orcarouter.ai/v1/chat/completions -H "Authorization: Bearer YOUR_ORCAROUTER_KEY" -H "Content-Type: application/json" -d '{ "model": "openai/gpt-4o-search-preview-2025-03-11", "messages": [{"role": "user", "content": "Hello"}], "max_tokens": 100 }'.
You can use all standard chat completions parameters supported by OpenAI: 'messages', 'max_tokens', 'temperature', 'top_p', 'frequency_penalty', 'presence_penalty', 'stop', and 'n'. The model supports a context window of 128,000 tokens, so 'max_tokens' can be set up to 16,384. 'temperature' controls randomness; values range from 0 to 2. 'stream' is supported for real-time response. OrcaRouter does not introduce additional parameters. Ensure your requests comply with the token limits; inputs longer than the context window (including the response) will be truncated or return an error.
Migrating is straightforward. Replace your OpenAI base URL (https://api.openai.com/v1) with OrcaRouter's base URL (https://api.orcarouter.ai/v1). Change your API key to an OrcaRouter key. Update the model field to openai/gpt-4o-search-preview-2025-03-11. No other code changes are required. If you were using a different OpenAI model, you can continue using the same code structure. OrcaRouter passes through all standard parameters. After migration, verify that responses are consistent. Note that billing will be handled by OrcaRouter, and you will need an account there.
OrcaRouter does not publicly disclose specific rate limits for this preview model. Availability depends on OpenAI's infrastructure and OrcaRouter's gateway capacity. For production applications, consider the model's preview status and potential instability. OrcaRouter may enforce reasonable usage limits to prevent abuse. You can expect best-effort service, but no SLA is provided. If you require guaranteed throughput, contact OrcaRouter for dedicated options. The model may also be removed or replaced without notice, so plan for fallback models.
The standard GPT-4o model (openai/gpt-4o) is a multimodal, stable release, while this preview is text-only and intended for early testing. The preview may include upcoming features that are not yet in the standard version, but it could also be less reliable. Pricing is similar. The standard model has broader input support (images, audio). If you do not need preview features, the standard GPT-4o is the safer choice. Both are available on OrcaRouter with the same API interface, making switching easy.
Gpt-4o-mini is a smaller, cheaper model optimized for simple tasks. It costs about $0.15 per million input tokens and $0.60 per million output tokens (approximate). This preview model is roughly 16x more expensive for input and 16x more expensive for output. The preview model offers larger context (128K vs. 128K? Actually gpt-4o-mini also has 128K, but that may vary. We know this model is 128K. Gpt-4o-mini's context is also 128K according to public info, but we can't assume. We'll say 'comparable context window' but note the price difference.) This model is better for complex reasoning, while gpt-4o-mini is sufficient for many standard tasks. Choose based on quality needs and budget.
Claude models (e.g., Claude 3.5 Sonnet) offer large context windows and strong reasoning, but their API is not directly OpenAI-compatible. OrcaRouter provides a unified interface for multiple providers, but this preview is exclusively from OpenAI. Claude models may have different pricing and strengths in areas like safety and long-form text generation. For direct comparison, you would need to test on your own use cases. OrcaRouter allows you to switch between providers, but this preview model is only available via the OpenAI route. No benchmark comparisons are provided.
OpenAI-compatible — keep the SDK you already use
https://api.orcarouter.ai/v1from openai import OpenAI
client = OpenAI(
base_url="https://api.orcarouter.ai/v1",
api_key="$ORCAROUTER_API_KEY",
)
response = client.chat.completions.create(
model="openai/gpt-4o-search-preview-2025-03-11",
messages=[{"role": "user", "content": "Hello"}],
)
print(response.choices[0].message.content)max_tokensresponse_formatstreamstructured_outputsweb_search_options| Input / 1M tokens | $2.50 |
| Output / 1M tokens | $10.00 |
| 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-search-preview-2025-03-11Open @misc{orcarouter_gpt_4o_search_preview_2025_03_11,
title = {openai/gpt-4o-search-preview-2025-03-11 API},
author = {openai},
year = {n.d.},
howpublished = {OrcaRouter},
url = {https://www.orcarouter.ai/models/openai/gpt-4o-search-preview-2025-03-11}
}openai. (n.d.). openai/gpt-4o-search-preview-2025-03-11 API. OrcaRouter. https://www.orcarouter.ai/models/openai/gpt-4o-search-preview-2025-03-11