I've been using Opus 4.5 for programmatic SEO and localizing game descriptions. If 4.6 truly improves context compaction, it could significantly lower the API costs for large-scale content generation. Has anyone tested its logic consistency on JSON output compared to 4.5?