💬 Reddit 4d ago
Major drop in intelligence across most major models.
User reports widespread quality degradation across major models (Claude, Gemini, Grok, z.ai) in mid-April 2026, observing ignored instructions, shallow outputs, and slow responses even when testing locally on H100 with GLM-5. Community discussion suggests potential systematic changes, though reports lack controlled verification. May reflect perception issues, A/B testing, or genuine model updates.