J.D. Hodges

Opus 4.7 launched on April 16, 2026. Here’s the fast path to switch Claude Code over, including the gotcha I hit in the VS Code extension.

claude update
claude --version
claude --model claude-opus-4-7

Step 1: Update Claude Code

Most people run Claude Code in a terminal. One command does it:

claude update

Anthropic lists v2.1.111+ as the minimum for Opus 4.7. If the /model slash command doesn’t recognize claude-opus-4-7 in your session, run claude update again and verify with:

claude --version

Step 2: Switch the model (pick one)

For the settings-file method, drop a model field into ~/.claude/settings.json:

{
  "permissions": { "allow": [] },
  "effortLevel": "max",
  "model": "claude-opus-4-7"
}

A project-level .claude/settings.json in a repo will override the global file for that project. Note that settings.json controls Claude Code’s runtime config, while your per-project CLAUDE.md still handles project instructions.

Running Claude Code inside VS Code? Use settings.json + panel reload

On release day, claude update moved me from v2.1.111 to v2.1.112, but the /model slash command still returned “Unknown command” inside my live VS Code panel. Here’s the path that actually worked:

1. Run claude update in any terminal.
2. Edit ~/.claude/settings.json and add the model field shown above.
3. Close the Claude Code panel in VS Code and reopen it. The new session picks up claude-opus-4-7 at launch.
4. Verify with the /status slash command, or just ask Claude which model it’s on.

Here’s the verification in a fresh VS Code session:

Why this works: the VS Code extension reads ~/.claude/settings.json when the panel launches. A panel that started before the new model existed won’t pick up a new model ID until it reloads.

What’s new in Opus 4.7

• Stronger coding: meaningful gains on agentic and terminal coding benchmarks (numbers below).
• New xhigh effort level, now the default for Opus 4.7 in Claude Code, giving finer control between high and max.
• Up to 2,576 px vision (~3.75 MP), roughly 3x prior resolution. Better accuracy on charts, dense documents, and screen UIs.
• /ultrareview: a dedicated code-review session in Claude Code. Pro and Max users get three free to try.
• Task budgets (separate feature, API public beta): cap token spending on long-running agent runs.
• Same API price as Opus 4.6: $5 per million input tokens, $25 per million output tokens. Opus 4.7 ships with a new tokenizer, so the same English input may map to roughly 1.0x to 1.35x more tokens than 4.6, which can nudge effective cost up slightly.

Opus 4.7 vs Opus 4.6 on the benchmarks that matter for Claude Code

Anthropic’s announcement has the full chart, including comparisons against GPT-5.4, Gemini 3.1 Pro, and the internal Mythos Preview. Worth a skim.

Note: On April 23, 2026, Opus 4.7 becomes the default for Enterprise pay-as-you-go and Anthropic API users. If you’re on one of those plans and happy to wait a week, the switch will happen automatically.

Sources: Introducing Claude Opus 4.7 (Anthropic), Claude Code model configuration docs, and hands-on verification in VS Code on release day.