Where does the bug appear (feature/product)?
Cursor IDE
Describe the Bug
With Max Mode enabled and “1M Max” explicitly selected in the model picker, the context window indicator in the Agent panel consistently shows a 300K token maximum (e.g. ~23.6K / 300K Tokens) instead of the expected 1M tokens.
This affects all models that advertise 1M context support — not just Claude Opus 4.6 but also Claude 4.5 Sonnet and others. The 300K cap appears to be a hard UI/backend limit regardless of which 1M-capable model is selected.
Per Cursor’s own documentation and pricing page, Max Mode should extend the context window to the model’s full maximum (1M for Opus 4.6 and Sonnet 4.5), with no long-context surcharge. The 1M context window for these models went GA in March 2026.
Steps to Reproduce
- Open Cursor IDE (v3.4.1 Nightly)
- Open the Agent panel
- Select any 1M-capable model (e.g. Opus 4.6)
- Enable Max Mode (toggle “1M Max” in model selector)
- Observe the context indicator — it shows
X / 300K Tokensinstead ofX / 1M Tokens - Repeat with other 1M models (Sonnet 4.5, etc.) — same 300K cap
Expected Behavior
The context indicator should show X / 1M Tokens (or ~1,000K) when Max Mode is enabled on a model that supports a 1M context window. The full 1M context should be usable before any summarization or condensation kicks in.
Operating System
MacOS
Version Information
Version: 3.4.1
VSCode Version: 1.105.1
Commit: 2a298dd06944a9b9ea541d28225b779fcbcc6200
Date: 2026-05-08T16:05:07.818Z
Layout: editor
Build Type: Stable
Release Track: Nightly
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 25.3.0
For AI issues: which model did you use?
Claude Opus 4.6 (with 1M Max enabled) — but the issue reproduces with all 1M-capable models.
Additional Information
This appears related to several other reports:
- Chat context limited to 70k for a 200K model — same class of bug but with different numbers
- Code-supernova-1-million model issue — 800K model stalling at 280K (~36%)
- Context limitation suddenly limits length — premature condensation
The pattern suggests Cursor has an internal cap (possibly 300K or lower) that overrides the model’s actual context window, even when Max Mode is toggled on and the UI shows “1M Max” in the model selector.
Does this stop you from using Cursor
Sometimes - I can sometimes use Cursor