Together AI: Large Reasoning Models Fail to Follow Instructions — New Benchmark Reveals Critical Flaw | SignalBreak | SignalBreak