Claude Opus 4.7: The Complete Guide to New Features and Specs

Key specs, new features (xhigh, Task Budgets, /ultrareview), performance gains, and tokenizer-change cautions for Claude Opus 4.7, released April 16, 2026.

🌐 This article was machine-translated and may contain inaccuracies. Refer to the Korean original if in doubt.

On April 16, 2026, Anthropic officially released Claude Opus 4.7. Its model ID is claude-opus-4-7, and it is currently the most powerful flagship among the publicly available Claude models.

Key specs

  • Context window: 1 million tokens (1M)
  • Max output: 128K tokens
  • Adaptive thinking: automatically adjusts reasoning depth based on task complexity
  • High-resolution vision: supports images up to roughly 3.75 megapixels (2,576px) — 3x the resolution of Opus 4.6
  • Pricing: $5 per million input tokens, $25 per million output tokens (same as Opus 4.6)

New features

xhigh effort level

A new xhigh level has been added between the existing high and max. It gives finer control over the trade-off between reasoning depth, cost, and speed. Claude Code uses xhigh as the default in many coding workflows.

Task Budgets

A feature for setting a token-budget cap on long-running agent tasks. The model treats the remaining budget as a countdown and wraps up gracefully when the budget runs out. Analysts called it the most underrated feature of launch week.

/ultrareview (Claude Code)

A multi-pass code-review command that examines bugs, edge cases, security issues, and logic errors more deeply than a standard review. Pro and Max subscribers get a set number of free uses per billing cycle.

Performance gains

Coding and agent tasks improved substantially. The SWE-bench Pro score rose about 10.9 points, from 53.4% on Opus 4.6 to 64.3%, and SWE-bench Verified rose from 80.8% to 87.6%.

Caution: tokenizer change

Opus 4.7 uses a new tokenizer. Even for identical input, the token count may increase by 1.0–1.35x (up to about 35%) depending on the content. Pricing itself is frozen, but actual token consumption may rise, so it's worth re-checking the cost estimates of your existing workflows. We recommend leaving headroom in the max_tokens parameter.

Available platforms

Available from launch day in the Claude app (Pro/Max/Team/Enterprise), the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, Claude Code, Cursor, and GitHub Copilot.


This article is based on public official announcements and press materials. Model specs and features may be updated; check Anthropic's official documentation for the latest.