[BUG] Claude Model Missmatch

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?

Bug Report: Model Picker Selection Not Honored — Opus 4.6 Selected, Sonnet 4.6 Responding
Product: Claude Code (VSCode/Cursor Extension)
Version: v2.1.47
Platform: macOS (darwin 25.2.0)
Plan: Claude Max (20x)
Date: 2026-04-10

Summary
When "Default (recommended) — Opus 4.6 with 1M context" is selected in the model picker, Claude Code consistently responds using claude-sonnet-4-6 instead of claude-opus-4-6. The model picker selection has no observable effect on the model actually serving responses.

Steps to Reproduce
Open Claude Code in Cursor/VSCode extension
Open the model picker
Confirm "Default (recommended) — Opus 4.6 with 1M context" is selected (checkmark visible)
Submit any prompt (e.g. "which claude model are we currently using?")
Observe the model ID in the response
Expected Behavior
Claude responds using claude-opus-4-6, consistent with the model picker selection.

Actual Behavior
Claude consistently responds using claude-sonnet-4-6 regardless of model picker selection. This was reproduced across:

Multiple separate conversations ("Check current Claude model version", "Verify database servers", "Standardize notification", "Change the AI model")
Multiple explicit queries ("which claude model are we currently using?", /model command)
Both before and after switching the model picker open/closed
Every response returned: Claude Sonnet 4.6 (claude-sonnet-4-6)

Evidence
Screenshot Observation
Model picker "Default (recommended) — Opus 4.6 with 1M context" selected with checkmark
Response (repeated) You're on Claude Sonnet 4.6 (claude-sonnet-4-6)
/stats output Sonnet 4.6 · Claude Max shown as active model
/model command Shows "Default (recommended)" — still responds with Sonnet
Terminal claude . Claude Code v2.1.47 — Sonnet 4.6 · Claude Max
Usage page "Sonnet only" weekly limit: 100% used. "All models": 91% used
Additional Context
Usage page breakdown (last screenshot) reveals a critical detail:

Current session 37% used (resets in 2hr 56min)
All models 91% used (resets in 8hr 56min)
Sonnet only 100% used (resets Sat 11:00 AM)
The "Sonnet only" sublimit is at 100% — but Claude is still responding with Sonnet, not falling back to Opus. This suggests one of two things:

Hypothesis A (routing bug): The extension ignores the model picker entirely and hardcodes Sonnet routing for all requests, regardless of selection or rate limits.
Hypothesis B (misleading UI): "Default (recommended)" does not mean "use Opus" — it means "Claude Code picks internally," and it picks Sonnet for most tasks. Opus is only invoked for specific complex task types (never triggered in this session). The UI label is misleading because it says "Opus 4.6" but behaves as Sonnet by default.
Either way: the UI is broken — it either lies about which model will be used, or it fails to route to the selected model.

What Was NOT the Issue
Plan access: User confirmed Opus access is included in their Max plan
The issue is not a missing Opus entitlement
The "Favorite model: Haiku 4.5" stat is unrelated — that is from internal subagent workers spawned automatically, not user-selected
Reproduction Rate
100% — every prompt in every conversation responded with Sonnet 4.6 despite Opus being selected.

What Should Happen?

Model Selector has to work. Currently selected model is not the one which is working.

Error Messages/Logs

Steps to Reproduce

Select -- Opus 4.6 ( Default )
Ask which model is in use. The response will be Sonnet 4.6
As well u will reach the week max amount of tokens used by sonnet withuor actually using Sonnet.
What exactly happened with me.

Claude Model

Opus

Is this a regression?

Yes, this worked in a previous version

Last Working Version

_No response_

Claude Code Version

v2.1.47

Platform

Anthropic API

Operating System

macOS

Terminal/Shell

Cursor

Additional Information

_No response_