I have been using gpt-4.1 model for ResponsesAPI to implement web search with OpenAI and it has been working properly. Once GPT-5.2 model is launched, I have upgraded the model to GPT-5.2. But it seems like it’s not enabled to do web search.
When asking “Can you do web search“ for GPT-4.1 model, it worked great but for GPT-5.2, it’s not working. Also, my agent is asking any additional data source instead of doing web search internally.
Is there a chance you are using “prompts” as a preset for holding tools, etc? You may need to create one from scratch or simply make full API requests, as problems already have been reported with this “instruction+tool” mechanism vs 5.2.
Here’s a test in the “playground”, not saving a prompt, using current version web search with a user location:
Further, this is a reasoning model, and needs reasoning.effort “low” or above to really search well (the default is “none”, not recommended): tool calls are internal to reasoning, and can have multiple uses in a different pattern than you are used to, taking much longer before output is produced and needing an increase in any timeouts.
