Opus 4.7 explained in 30 seconds

My notes

Summary

A 30-second TikTok rundown of Claude Opus 4.7: measurable gains on the three big coding benchmarks, 3x higher screenshot resolution handling, higher token usage at equivalent effort levels, a new “X high” (extra high) reasoning effort tier (now default, one step below max), and a new /ultra-review slash command that opens a dedicated code review session. Trade-off: better performance at higher cost.

Key Insight

• Coding benchmark jump. 4.7 improves on all three major coding evals vs 4.6, described as a solid jump forward purely on numbers.
• Vision bump: 3x screenshot resolution. Meaningful for UI review, design critique, screenshot-driven debugging, OCR-like tasks. Previous limits forced downscaling; 4.7 reads more detail per image.
• Token hog warning. At the same effort level, 4.7 burns more tokens than 4.6. Per-task cost goes up even when settings stay the same, budget accordingly.
• New effort tier: X high (extra high). Sits between “high” and “max”. Now the default. Upgrading from 4.6 defaults without changing effort means both better output and a visible cost bump.
• New /ultra-review slash command. Dedicated review session for a project, distinct from regular chat or code edits. Likely positioned as the go-to for pre-commit or pre-PR review passes.
• Net framing: 4.7 is not a Pareto improvement. It’s better output at higher cost. Worth it for high-stakes coding and review; probably overkill for routine edits where 4.6 at lower effort is still sensible.