[BUG] Cowork forces claude-opus-4-6[1m] (1M context) causing instant rate limit errors on Max plan

Preflight Checklist

• [x] I have searched existing issues and this hasn't been reported yet
• [x] This is a single bug report (please file separate reports for different bugs)
• [x] I am using the latest version of Claude Code

What's Wrong?

After upgrading Claude Desktop on Mac, the Cowork tab immediately returns "API Error: Rate limit reached" on every request — including a simple "hello" prompt — despite being on the Max plan with available quota. Chat tab works fine.

Root cause identified in logs: the new Cowork version passes --model claude-opus-4-6[1m] (1M context window flag) for all sessions. The 1M context window has significantly tighter rate limits, causing instant rate limiting even on Max. The previous working version used --model claude-opus-4-6 (standard context) and worked perfectly.

Additionally, Cowork spawns 6-12 sub-agent processes for a single trivial prompt, each requesting [1m] context. When Opus [1m] fails, it cascades to claude-sonnet-4-6[1m] then claude-haiku-4-5-20251001 — all hitting rate limits.

What Should Happen?

Cowork should either use standard context (claude-opus-4-6) by default instead of [1m], or the 1M context rate limits should be compatible with Max plan quotas. A simple "hello" prompt on Max plan should not be rate-limited.

Error Messages/Logs

Rate limit errors from `main.log`:


2026-03-11 02:46:32 [info] [APIError] Intermediate SDK error "rate_limit" for session local_d4383f34-9611-4db5-999f-2c372ff6add9 — continuing to read stream
2026-03-11 02:46:32 [info] [Result] Turn failed for session local_d4383f34: subtype=success, is_error=true
  unhealthy_reason: 'api_error',
  error_message: 'API Error: Rate limit reached',

2026-03-11 02:49:48 [info] [APIError] Intermediate SDK error "rate_limit" for session local_d5cf88b3-9634-4aff-9889-987006385296
2026-03-11 02:50:17 [info] [APIError] Intermediate SDK error "rate_limit" for session local_ad8ebe1e-3ea9-42ca-9223-77aed2fd9f61


Broken sessions from `cowork_vm_node.log` — note `[1m]` flag:


2026-03-11 02:26:08 [Spawn:create] --model claude-opus-4-6[1m]
2026-03-11 02:27:33 [Spawn:create] --model claude-opus-4-6[1m]
2026-03-11 02:28:25 [Spawn:create] --model claude-opus-4-6[1m]
2026-03-11 02:28:32 [Spawn:create] --model claude-opus-4-6[1m]
2026-03-11 02:29:45 [Spawn:create] --model claude-opus-4-6[1m]
2026-03-11 02:31:04 [Spawn:create] --model claude-opus-4-6[1m]


Working session from older version — no `[1m]` flag:


2026-03-06 23:09:19 [Spawn:create] --model claude-opus-4-6


Additional errors:


[warn] [startVM] Post-connect setup failed: 'RPC error: failed to write SDK binary: open /usr/local/bin/claude: text file busy'
[info] [VM:network] Not connected, will show error in 30000ms if not resolved

Steps to Reproduce

1. Have Claude Max plan ($200/month)
2. Update Claude Desktop to latest version (v1.1.6041)
3. Open the Cowork tab
4. Type "hello" and send
5. Immediately get "API Error: Rate limit reached"
6. Verify Chat tab works fine on the same account with the same prompt

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

Claude Desktop v1.1.4173 (app_version from main1.log, dated Feb 25, 2026). This version used --model claude-opus-4-6 without the [1m] flag. Downgrading to v1.1.4088 also works.

Claude Code Version

Claude Desktop v1.1.6041 (broken) CCD version 2.1.51 (from main.log: "Initialized with version 2.1.51")

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Other

Additional Information

Workaround: Downgrading to v1.1.4088 and disabling auto-updates resolves the issue:

defaults write com.anthropic.claude SUEnableAutomaticChecks -bool false

Key finding: The regression is the [1m] context flag being appended to model names in Cowork sessions. Comparing cowork_vm_node.log entries:

• v1.1.4173 (working): --model claude-opus-4-6
• v1.1.6041 (broken): --model claude-opus-4-6[1m]

Suggested fixes:

1. Make [1m] context opt-in rather than default for Cowork
2. Adjust 1M context rate limits to work within Max plan allocations
3. Add a model/context selector in the Cowork UI

Possibly related issues: #28975, #29579, #27336, #28832