Anthropic valued at $965B with a confidential IPO, Opus 4.8 with 1,000 sub-agents, Gemini Spark 24/7: the week everything tipped over.
Opus 4.7 dropped on April 16. Most devs use it exactly like Sonnet. Effort mode, cache, vision, instruction following, routing: 5 specific settings, with use cases and pitfalls.
Anthropic released Claude Opus 4.7 today. Box cut AI costs by 30%, XBOW's penetration testing jumped from 54% to 98.5%, and the benchmark table publicly shows a better model you can't access yet.
Since February, developers have been saying Claude Opus 4.6 got worse. I went back through the tweets, GitHub issues, and Anthropic replies to understand what actually happened.
€20 in revenue over a year, €40/month in infrastructure. I'm shutting down Livate. Here are the raw numbers, the reasons, and what I'm building next.
Not with /compact. With what I set up BEFORE opening a session. CLAUDE.md, hooks, and proactive rules: my setup to cut token consumption by 40 to 60%.
