Cursor AI not providing model I chose

Describe the Bug

Cursor AI is not providing me with the model I chose. When I select Claude 4 Sonnet, and ask which model it is, it lies that it is Claude 4, but its cutoff time is April 2024. When interrogated about this, it admits that its Claude 3.5 Sonnet.
I removed GPT 5 model for now, because its very slow, and it doesnt provide me with “Keep changes” option, yet in Auto mode, it constantly selects GPT 5

Steps to Reproduce

Select Claude 4 Sonnet, remove Sonnet 3.5 from model list. Ask which model it is and check the result. Interrogate about the cutoff day.
For GPT 5, do the same

Expected Behavior

I expect to get the model I chose, and consequently, to not get GPT 5 when I remove it from model list/toggle it off

Screenshots / Screen Recordings

Operating System

Windows 10/11

Current Cursor Version (Menu → About Cursor → Copy)

Version: 1.4.3 (user setup)
VSCode Version: 1.99.3
Commit: e50823e9ded15fddfd743c7122b4724130c25df0
Date: 2025-08-08T17:34:53.060Z
Electron: 34.5.1
Chromium: 132.0.6834.210
Node.js: 20.19.0
V8: 13.2.152.41-electron.0
OS: Windows_NT x64 10.0.19045

Does this stop you from using Cursor

No - Cursor works, but with this issue

LLMs can’t know what models they are
so its not true

This is a common issue/misconception with all LLMs. Even on Anthropic’s website, Claude will report the wrong model!

In short the LLM typically doesn’t know it’s model number or report it’s training data date correctly. Cursor is not going to route your request to the wrong selected model.

Note, in case it’s relevant, Auto doesn’t care about your model selection - it will always route to whatever one Cursor deems appropriate at the time.

1 Like

Alright, thank you for the explanation. However, shouldnt it at least know its knowledge cutoff time?

At least this puts me to rest for Claude 4. GPT 5 is really slow and unmethodical currently in Cursor. I assume that will change, but..its not good

Nah they’ve always been ■■■■ at knowing the cut off time.

Claude will even fight with you, considering your data as the ‘future’ and hardcode time data to compensate!

This topic was automatically closed 22 days after the last reply. New replies are no longer allowed.