This browser is no longer supported.
Upgrade to Microsoft Edge to take advantage of the latest features, security updates, and technical support.
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(57.599999999999994, 27%, 15%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3ERW%3C/text%3E%3C/svg%3E)
I have added an Azure Search Service index with the options for semantic configuration and a Vector Profile. When I execute the following REST request. The following error is raised. Does anyone have some insight into this result?
[Error Message] An error occurred when calling Azure OpenAI: Server responded with status 400. Error message: {
"error": {
"message": "Unsupported parameter: 'max_tokens' is not supported with this model. Use 'max_completion_tokens' instead.",
"type": "invalid_request_error",
"param": "max_tokens",
"code": "unsupported_parameter"
}
}
As you can see, I make no reference to max_tokens. When I add max_completion_tokens, reasoning_effort, or verbosity, an error is raised indicating that these options are not supported.
My JSON Request
{
"messages": [
{
"role": "system",
"content": "My prompt here"
},
{
"role": "user",
"content": "show a bar chart"
}
],
"stop": null,
"stream": false,
"frequency_penalty": 0,
"presence_penalty": 0,
"temperature": 1,
"data_sources": [
{
"type": "azure_search",
"parameters": {
"endpoint": "https:\/\/....search.windows.net",
"index_name": "aisearch-indexset-42984f3678dd446fa51ee376e792c125",
"query_type": "semantic",
"semantic_configuration": "semantic-config",
"strictness": 3,
"top_n_documents": 10,
"in_scope": true,
"filter": "search.ismatch('LIB00000000000000*', 'dbdefid')",
"fields_mapping": {
"content_fields_separator": "\\n",
"content_fields": [
"content"
],
"filepath_field": "dxlink",
"title_field": "exportfilename",
"url_field": "dxlink",
"vector_fields": [
"vector_content"
]
},
"authentication": {
"type": "api_key",
"key": "...."
}
}
}
]
}
Thank you in advance.
Ron
Noted. We have also used this parameter and others and were sent errors that those parameters were not accepted either.
We are only guessing here but I need to confirm that GPT-5 supports RAG.
Thx
Ron
![](data:image/svg+xml, %3Csvg xmlns='http://www.w3.org/2000/svg' height='64' class='font-weight-bold' style='font: 600 30.11764705882353px "SegoeUI", Arial' width='64'%3E%3Ccircle fill='hsl(32, 9%, 13%)' cx='32' cy='32' r='32' /%3E%3Ctext x='50%25' y='55%25' dominant-baseline='middle' text-anchor='middle' fill='%23FFF' %3EAK%3C/text%3E%3C/svg%3E)
Yes, GPT-5 supports and significantly enhances RAG. In-fact the general consensus in the AI community is the GPT-5 is a major advantage for RAG applications.
Here’s why:
1. Enhanced Reasoning and Grounding: GPT-5 models are reported two have superior reasoning capabilities and are specifically aligned to reduce hallucination this is a critical for RAG, as the model is better at analyzing the retrieved documents and producing an answer that is strictly grounded in that information.
2. Improved Tool Use: GPT-5 shows significant gains in benchmarks related to instructions following an agentic tool use. RAG is fundamentally an external tool the model uses to retrieve data. Better tool use means that model is more reliable in deciding when to use the retrieval and how to synthesize the retrieved information effectively.
3. High-Context Window: While not the only factor, a larger context window makes it easier to pass a greater volume of retrieve documents to the model, giving it more information to generate an accurate response.
4. Specific Deployment Support: As shown in the images you provided, user actively trying to implement RAG with GPT 5 on the platforms like Azure open AI service which is primarily method for enterprise level RAG integration using tools like Azure AI search
The reason you don't see connect dataset option for GPT-5 itself is because RAG is a technique that is applied to the model, not a feature built into the model's core architecture.
The GPT-5 model is the massive, pre-trained neural network and RAG is a system architecture built around the core GPT-5 model to overcome the knowledge cutoff and hallucination issues. The main reason of not having the option is security & scalability and also decoupling.
Then why is it possible to use GTP-4o and not GPT-5
the difference is likely an evolution of platform's user experience and architecture: what was an explicit "option" you clicked for GPT-4 has become more seamless, integrated and decoupled architectural component in the GPT-5 ecosystem, perhaps only visible to system administrators or via advanced API configuration rather than a simple user interface checkbox.
So we are back to my original question why is this json request not working with GPT-5.
Please refer to these documentations for better understanding:
"Hope you're doing well. We're following up on the ticket. Please let us know before 15 October whether the issue persists or has been resolved, after that we'll close this thread and If the issue has been resolved, kindly mark the response as answered."
"Hope you're doing well. We're following up on the ticket. Please let us know before 15 October whether the issue persists or has been resolved, after that we'll close this thread and If the issue has been resolved, kindly mark the response as answered."
We haven’t received any further updates or follow-up on this case, so we’ll proceed with closing it at this time. If you require additional assistance or have new information to share, please feel free to raise new thread, we’re happy to support you further needed.
My original questions was never answered. If your unable to provide an answer you can close the request.