VS Code Codex reports 950k context but compacts/fails around 260k-270k

What version of the IDE extension are you using?

vs code 1.125.0

What subscription do you have?

Pro (200)

Which IDE are you using?

VS Code 1.125.0

What platform is your computer?

Windows 11

What issue are you seeing?

Title: VS Code Codex reports 950k context but compacts/fails around 260k-270k

VS Code Codex appears to report a high configured/effective context window (950000) but still auto-compacts or fails around the old ~260k-270k input-token boundary.

This looks like a split-brain effective-context issue: one path reads the configured large window, while the compaction or backend rejection path still behaves as if the active window is based on 272000 * 95% = 258400.

Environment:

• OS: Windows 11 Pro
• Platform string: Microsoft Windows NT 10.0.26200.0 x64
• IDE: VS Code
• Subscription: ChatGPT Pro ($200/month)
• VS Code extension: openai.chatgpt 26.5616.31447
• VS Code bundled Codex runtime: codex-cli 0.142.0-alpha.1
• Terminal Codex CLI: codex-cli 0.141.0
• Rollout originator: codex_vscode
• Model: gpt-5.5
• Windows 11 Pro / Microsoft Windows NT 10.0.26200.0 x64

Local ~/.codex/config.toml includes:

model = "gpt-5.5"
model_context_window = 1000000
model_auto_compact_token_limit = 900000

### What steps can reproduce the bug?

Any conversation longer than ~260K simply gets auto-compated

### What is the expected behavior?

_No response_

### Additional information

_No response_