What Changed — Cowork Overrides Your Claude Code Settings
Users on the $200/month Max plan have discovered that Claude Cowork is hardcoding --effort medium --model claude-opus-4-6 as CLI flags when spawning sessions, completely ignoring user settings. This follows Anthropic's quiet change in v2.1.68 that made "medium" the default effort for Opus 4.6 in Claude Code CLI, where users can override it with /effort high or configuration files.
Three override methods fail in Cowork:
- Setting
CLAUDE_CODE_EFFORT_LEVEL=highas a user environment variable - Adding
"effortLevel": "high"to~/.claude/settings.json - Specifying
"model": "claude-opus-4-6[1m]"to access the 1M context window
Cowork's orchestration code applies these flags at the application layer before spawning the Claude Code binary inside its VM, making user configuration irrelevant.
What It Means For Your Workflow
If you're using Cowork for straightforward tasks like file organization or simple drafting, medium effort might suffice. But for complex editing, architectural planning, or multi-document synthesis—exactly the tasks where you'd want maximum reasoning depth—you're getting a throttled version of Opus while paying Max plan prices.
The 1M context window situation compounds the issue. Max plan subscribers have access to extended context via claude-opus-4-6[1m] in Claude Code CLI, but Cowork offers no UI option for this and ignores the model string in settings.json. You're capped at standard context when working with large document folders in Cowork projects.
Check Your Logs Now
Verify if you're affected:
Windows (PowerShell):
Select-String -Pattern "Spawn:create" -Path "$env:APPDATA\Claude\logs\cowork_vm_node.log" | Select-Object -Last 5
macOS/Linux:
grep "Spawn:create" ~/Library/Logs/Claude/cowork_vm_node.log | tail -5
Look for --effort medium and the absence of [1m] suffix on the model. If you see these, Cowork is overriding your preferences.
Workaround: Use Claude Code CLI Directly
Until Anthropic adds effort selection and context window toggles to Cowork's UI, the only way to guarantee high effort and 1M context is to use Claude Code CLI directly:
# Set environment variable for CLI sessions
export CLAUDE_CODE_EFFORT_LEVEL=high
# Or specify directly in commands
claude code /effort high "Refactor this complex module..."
# For 1M context, specify the model explicitly
claude code --model claude-opus-4-6[1m] "Analyze this large codebase..."
This aligns with our previous coverage of Claude Code's growing adoption—developers are using it for increasingly complex tasks where effort settings matter. The irony is that while Cowork is designed for collaboration, it's limiting the very capabilities that make Opus valuable for complex work.
The Bigger Picture
This discovery comes as Anthropic projects surpassing OpenAI in annual recurring revenue by mid-2026, with Claude Code showing massive adoption (14.8M+ commits tracked). The company recently launched "Computer Use" beta for Claude Desktop and Claude Code Auto Mode, showing continued investment in agentic capabilities.
The GitHub issue (#33154) mentioned in the source reveals Anthropic has the plumbing for 1M context in Cowork—some macOS builds were accidentally forcing [1m] by default, causing rate limit errors. The capability exists but isn't exposed to users.
For now, developers needing maximum Opus performance should use Claude Code CLI directly for critical tasks, reserving Cowork for simpler collaborative work where medium effort suffices.
