DeepSeek V4: context limited to 200K + reasoning_content error

Where does the bug appear (feature/product)?

Cursor IDE

Describe the Bug

Issue 1: Context window capped at 200K
DeepSeek V4 Pro has a native 1M token context window. When connected through Cursor, the context appears to be limited to only 200K tokens. Other tools (e.g., OpenCode) pass through the full 1M context without restriction. Please support the full 1M context window.
Issue 2: “reasoning_content must be passed back to the API” error
When DeepSeek V4’s thinking mode is enabled, Cursor throws:
{“error”:{“message”:“The reasoning_content in the thinking mode must be passed back to the API.”,“type”:“invalid_request_error”}}
DeepSeek’s API requires the reasoning_content from the previous response to be included unmodified in subsequent requests. Cursor’s agent loop does not preserve/relay this field, causing all follow-up requests to fail. Please implement proper reasoning_content passthrough for DeepSeek models.

Steps to Reproduce

Issue 1:

  1. Connect DeepSeek V4 in Cursor
  2. Load a codebase larger than 200K tokens
  3. Observe that only ~200K tokens of context are available, even though the model supports 1M
    Issue 2:
  4. Enable thinking/reasoning mode for DeepSeek V4 in Cursor settings
  5. Send a multi-turn conversation (2+ messages)
  6. Error occurs on the second request: “reasoning_content must be passed back to the API”

Operating System

MacOS

Version Information

Version: 3.2.11
VSCode Version: 1.105.1
Commit: e9ee1339915a927dfb2df4a836dd9c8337e17cc0
Date: 2026-04-24T14:36:47.933Z
Layout: editor
Build Type: Stable
Release Track: Default
Electron: 39.8.1
Chromium: 142.0.7444.265
Node.js: 22.22.1
V8: 14.2.231.22-electron.0
OS: Darwin arm64 25.4.0

For AI issues: which model did you use?

DeepSeekV4pro

For AI issues: add Request ID with privacy disabled

Request ID: 08ff07d7-7821-4f23-8329-4c3076ce4aab
{“error”:“ERROR_PROVIDER_ERROR”,“details”:{“title”:“Provider Error”,“detail”:“Provider returned error: {"error":{"message":"The reasoning_content in the thinking mode must be passed back to the API.","type":"invalid_request_error","param":null,"code":"invalid_request_error"}}”,“isRetryable”:false,“additionalInfo”:{},“buttons”:,“planChoices”:},“isExpected”:true}
Provider Error Provider returned error: {“error”:{“message”:“The reasoning_content in the thinking mode must be passed back to the API.”,“type”:“invalid_request_error”,“param”:null,“code”:“invalid_request_error”}}
Cvs: Provider Error Provider returned error: {“error”:{“message”:“The reasoning_content in the thinking mode must be passed back to the API.”,“type”:“invalid_request_error”,“param”:null,“code”:“invalid_request_error”}}
at Pf_ (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:29432:24260)
at Rf_ (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:29432:23144)
at Uf_ (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:29433:6490)
at q0h.run (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:29433:11307)
at async cQi.runAgentLoop (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:41748:12418)
at async Vnm.streamFromAgentBackend (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:41816:12344)
at async Vnm.getAgentStreamResponse (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:41816:19567)
at async lCt.submitChatMaybeAbortCurrent (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:29548:16906)
at async Va (vscode-file://vscode-app/Applications/Cursor.app/Contents/Resources/app/out/vs/workbench/workbench.desktop.main.js:40804:4230)

Does this stop you from using Cursor

No - Cursor works, but with this issue

7 posts were merged into an existing topic: Compatibility with DeepSeek model’s design to return reasoning content after tool calls?