reasoningfactbearishLarge audio-language models still lag behind text-based large language models in deep logical reasoning, primarily due to the scarcity of high-quality audio reasoning dataMachine Learning29 Jul 2026http://arxiv.org/abs/2607.21550v1