Anthropic says its new AI model Claude Opus 4.7 is less likely to hallucinate and engage in sycophancy than most other models.