Claude Opus 4.7 Is Coming This Week: Should You Upgrade Your OpenClaw?

The Information reported today that Anthropic is preparing Claude Opus 4.7 for release as early as this week. It's a distinct model from Claude Mythos (the cybersecurity-focused frontier model) — this is the next general-purpose Opus. Here's what we know, how to upgrade OpenClaw when it drops, and the one thing to check before you switch.

What We Know About Opus 4.7

Update: Opus 4.7 launched today, April 16, 2026. Here's what's confirmed:

• API model ID: claude-opus-4-7
• Context window: 1M tokens (same as 4.6)
• Max output: 128K tokens
• High-res image support: Up to 2576px / 3.75MP — up from 1568px. Big deal for screenshot, document, and computer-use workflows.
• New xhigh effort level: Extra intelligence at the cost of more tokens. Use for complex reasoning tasks.
• Distinct from Mythos. Mythos is Anthropic's specialized cybersecurity model in gated preview. Opus 4.7 is the general-purpose model for OpenClaw.
• Caveat: OpenClaw's model catalog doesn't recognize it yet (issue #67710). See our fix guide before switching.

How to Upgrade OpenClaw When It Drops

When Opus 4.7 becomes available through the Anthropic API, it will surface as a new model identifier. The upgrade path in OpenClaw is straightforward:

Option A — Change the model string in openclaw.json

// Before
{
  "model": "claude/opus-4-6"
}

// After (update to whatever Anthropic names it)
{
  "model": "claude/opus-4-7"
}

Then: openclaw gateway restart

Option B — Use the configure command

openclaw configure

Navigate to the model selection screen and pick the new model from the list. OpenClaw pulls available models from the Anthropic API, so Opus 4.7 will appear automatically once it's live.

Option C — Per-agent override (test first)

If you want to test Opus 4.7 on a specific agent before switching everything:

// openclaw.json — test on one agent first
{
  "agents": {
    "test-agent": {
      "model": "claude/opus-4-7"
    }
  }
}

What to Test Before Going All-In

Every major model upgrade can change behavior in subtle ways. Before switching your primary agent, run a few validation checks:

1. SOUL.md compatibility

New models sometimes interpret persona instructions differently. Run a few turns with your SOUL.md loaded and check that tone, response length, and decision-making style feel right. Edge cases: explicit instruction-following, refusal thresholds, verbosity.

2. Tool call reliability

Test your most-used tool combinations — especially exec, web_search, and any custom skills. Model upgrades occasionally affect tool call formatting or multi-step tool chains.

3. Memory and context behavior

Run a few turns that require your agent to reference memory files and check it's pulling context correctly. If you're using LanceDB or memory-lancedb, verify retrieval still works as expected.

4. Cost per turn

New frontier models often have higher per-token pricing. Check Anthropic's pricing page immediately after launch and compare input/output costs to your current 4.6 spend before making a permanent switch.

Should You Upgrade Immediately?

If your current setup is running well, wait 24–48 hours after launch. Let the early adopters find the edge cases first. If you're already experiencing Opus 4.6 degradation (slower responses, unexpected stops, token creep), upgrading promptly makes sense — you have less to lose and potentially a lot to gain.

Watch For: Claude Code Routines

Alongside Opus 4.7, Anthropic launched Claude Code Routines — scheduled and event-triggered agents that run on Anthropic's managed cloud infrastructure. Think of it as cron jobs powered by Claude Code, without needing your own server.

For OpenClaw users, this is mostly complementary rather than competitive: OpenClaw's heartbeat system does similar things but runs on your own hardware with your own memory and full file/exec access. Routines are appealing for users who don't want to self-host but they lack the workspace persistence and channel flexibility that makes OpenClaw powerful.

We'll publish a full comparison once Claude Code Routines is out of beta.