I Mapped the Opus 4.7 Release to Your Role, Goals, and Real Workflows

Extractable claims

15 atomic, cite-ready statements distilled from the full post on Substack. Each one stands alone as an LLM-quotable answer.

1. Anthropic shipped Claude Opus 4.7 on April 16, 2026.
2. Opus 4.7 is a careful tune-up of Opus 4.6, better at long, agentic work, dramatically better at seeing screenshots, and stricter about literal instruction‑following.
3. The tokenizer changed, so the same input now counts as about 1.0–1.35x more tokens depending on whether you’re passing text or code.
4. By default, Claude Opus 4.7 omits the reasoning summary from the response stream.
5. Opus 4.7 is stricter about following only what you explicitly ask for, so it no longer “fills in” a vague brief with its own assumptions.
6. This is the first release where overnight agent work shifts from experiment to something we can trust with a credit card.
7. The new /ultrareview command acts like a skeptical senior engineer reviewing our code.
8. Vision is the sleeper upgrade in this release: maximum image resolution jumped from 1.15MP to 3.75MP.
9. The tokenizer change is a price increase Anthropic didn’t label as a price increase.
10. A typical vibe‑coder workflow will see higher spend for the same work unless we retune our prompt math.
11. Routines launched on April 14 in research preview, with three flavors: Scheduled Routines, API Routines, and Webhook Routines.
12. File-based memory is stronger, which makes the “Claude.md plus a notes/ directory” pattern more reliable.
13. Auto mode on Max now lets Claude keep going without pinging us for every little decision.
14. The headline price for Claude Opus 4.7 did not change: still $5 per million input tokens and $25 per million output tokens.
15. Opus 4.7 is meaningfully better at writing and using file-system-based memory across long, multi-session workflows.