Same failure mode has been reported across the GPT-5 line on the Responses API for months — looks like one persistent serving-side regression, not anything prompt-specific:
- GPT-5.2 streaming: two or more messages in same response stream — OpenAI confirmed users were billed for the erroneous output, then couldn’t reproduce.
-
- Intermittent duplicate strings in GPT-5.4 responses — staff asked for a repro, still open.
-
- Responses API: duplicate output messages + tool calls and duplicate response + invalid JSON — gpt-4.1 on the Responses API, since mid-2025.
Happy to provide request IDs for our gpt-5.5 doubled-vs-clean cases.
- Responses API: duplicate output messages + tool calls and duplicate response + invalid JSON — gpt-4.1 on the Responses API, since mid-2025.