EfficientApartment52 shares their experience and concerns about GitHub Copilot models being limited to 128k context length, inviting other users and the Copilot team to discuss the impact and potential solutions.

Capped Context Length Issues in Copilot - Anyone Else Experiencing This?

Author: EfficientApartment52

I’ve been testing various models in Copilot and noticed they’re all capping out at around 128k context length, even though some models like GPT-5 are supposed to handle up to 400k tokens. This limitation was discovered through debugging, and it causes ongoing conversations to get summarized too early, impacting continuity—especially for longer tasks and detailed threads. The same cap was observed with models like Sonnet-4, gemini-2.5-pro, and GPT-4.1.

Open Questions:

Has anyone else experienced similar limits?
Is this a known restriction or potentially a misconfiguration?

Concerns Raised:

Reduced utility of Copilot in long or complex conversations.
Premium requests seem to be less valuable if the context window is smaller than advertised.
Desire for greater transparency from the Copilot team regarding these operational limits.

Additional Points:

Screenshots are provided in the original post showing the detected context length limits for the models in question.
The author notes a recent switch to Copilot from Cursor, which indicates model context windows in the chat.
There’s speculation that context lengths are intentionally capped due to subscription and pricing tiers.

Community Call to Action: Anyone with similar experiences or information is invited to weigh in. The author also appeals to the Copilot team to address or clarify the actual supported context lengths for each model, and requests that limitations be fully communicated within the product interface.

For more discussion or screenshots, see the original Reddit thread.

This post appeared first on “Reddit GitHub”. Read the entire article here