claude-code - 💡(How to fix) Fix Sonnet 4.6 routes every request to long-context tier on Claude Desktop 2.1.149 (429 'Usage credits required')

Summary

On Claude Desktop 2.1.149 (Windows), every request sent to claude-sonnet-4-6 is routed to the long-context (1M) billing tier, even on fresh sessions with very little context (~46K tokens). The API returns 429 rate_limit_error: "Usage credits are required for long context requests." and Sonnet 4.6 becomes unusable on the subscription plan.

Switching the same session to claude-opus-4-7 works fine — Opus dispatches successfully against the standard tier.

Environment

• Claude Desktop: 2.1.149 (entrypoint: claude-desktop)
• Agent SDK: 0.3.149
• OS: Windows 11 Pro 10.0.26200
• Plan: subscription (no API credits purchased)
• No CLAUDE_CODE_* 1M-context env vars set
• ~/.claude/settings.json is empty ({})
• Project settings only contain a single Bash permission allow — nothing context-related

Reproduction

1. Open a fresh session in Claude Desktop on claude-sonnet-4-6.
2. Send any small message (session well under 200K tokens — observed at ~46K).
3. Request fails with 429.

Observed behavior

From the debug log (claude --debug):

[DEBUG] autocompact: tokens=[REDACTED] level=ok effectiveWindow=980000
[DEBUG] [API:timing] dispatching to firstParty model=claude-sonnet-4-6
[DEBUG] [API REQUEST] /v1/messages x-client-request-id=28f57bba-c879-402b-a0e2-34e7bd75a9df source=sdk
[ERROR] API error (attempt 1/11): 429 {"type":"error","error":{"type":"rate_limit_error","message":"Usage credits are required for long context requests."},"request_id":"req_011CbQ1dZ1fxiLrtY5LDScQv"}
[ERROR] API rate_limit after retries: Usage credits are required for long context requests.

Key signal: effectiveWindow=980000 — the client appears to be advertising a ~1M-token window by default for Sonnet 4.6, which pushes every request (regardless of actual token count) into the long-context tier that requires prepaid API credits.

In the same debug session, requests dispatched to claude-opus-4-7 succeed normally, which suggests the 1M-window default is being applied selectively to Sonnet 4.6.

Expected behavior

On the subscription plan, Sonnet 4.6 requests under 200K tokens should route to the standard tier and succeed, matching Opus 4.7's behavior. The 1M long-context tier should only be engaged when context actually exceeds the standard window.

Request IDs for server-log lookup

• req_011CbQ1dZ1fxiLrtY5LDScQv
• req_011CbQ2825NmrtGRDvwZSdXW
• req_011CbQ28ARBPJDZ6TDN1EruQ

Workaround

Switch model to claude-opus-4-7 via /model claude-opus-4-7. Sonnet 4.6 remains unusable until this routing behavior is corrected or the client stops defaulting to the 1M window.