A subagent with 1M context starts with 272k context

Support Bug Reports
, ,
1

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Subagent Luna-5.6-Max-1M compresses context too early for the task it was launched for.

Steps to Reproduce

---
name: Verifier Pro
model: gpt-5.6-luna[context=1m,reasoning=max,fast=false]
description: Verifier Pro
---

Expected Behavior

1M model can audit 2k dif LOC without summarizing.

Screenshots / Screen Recordings

2026-07-12_22-26-15.png
2026-07-12_22-26-15.png913×845 22.6 KB

Operating System

Windows 10/11

Version Information

Version: 3.11.18 (system setup)
VS Code Extension API: 1.125.0
Commit: a1d1c20f31cb834d7d50581e6da017d2b822e380
Date: 2026-07-11T21:40:27.094Z
Layout: IDE
Build Type: Stable
Release Track: Nightly
Electron: 40.10.3
Chromium: 144.0.7559.236
Node.js: 24.15.0
V8: 14.4.258.32-electron.0
xterm.js: 6.1.0-beta.256
OS: Windows_NT x64 10.0.22631

For AI issues: which model did you use?

Grok-4.5-High + GPT-5.6-Luna-Max-1M

For AI issues: add Request ID with privacy disabled

Agent: 3be29201-b22c-49c1-8428-f87972b3792e
Subagent: daea9695-be20-4799-b369-323130f4d7e5

Does this stop you from using Cursor

Yes - Cursor is unusable

5

Hi @Artemonim!

Thanks for the report.

I think I know what this bug is — could you toggle “Max Mode” on and see if that changes the behavior?

7

Yes, that sounds like a reason.

2026-07-13_15-45-23
2026-07-13_15-45-231180×519 19.5 KB

Is this how it is intended or will it be corrected?

9

With the impending removal of Max Mode for usage-based plans, I imagine this will need to be corrected! I’ll still file a bug report for the team, however, as I can reproduce the issue myself.

10

@Colin I’m sorry for the off-topic, but could you also take a look at my issue in the GPT-5.6 feedback thread?