Has anyone experience a degradation in parallel tool calling and parallel agent dispatch with GPT-5 series models? the parallel_tool_call param is confirmed to be True, the dispatch is async, and the reasoning effort doesn’t matter, it just won’t do it. GPT 4.1, Claude models, etc all do it fine.
Would love to hear your experience and whether you noticed a regression, found a workaround, or just want to vent!