Did any of you experience any kind of depreciation in the overall code generation for sonnet 4 in copilot?

[KrishGaur1354]

Select Topic Area

Question

Copilot Feature Area

VS Code

Body

Anthropic did admit it: https://status.anthropic.com/incidents/72f99lh1cj2c

[jfullstackdev]

it was just in August I encountered Claude Sonnet 4, and it was a different experience than the 3.7, so I don't know what you refer to as "kind of depreciation"

if it was not too heavy consuming the Premium Request, that would be my day to day, unfortunately, its power comes with that trade-off, so I default to GPT-4.1 as may base, and only needing Claude Sonnet 4 for the initial planning or for heavy tasks

[FrandyAquino]

ChatGPT said:

Absolutely — you weren’t imagining things 👀.

Anthropic confirmed in their incident report
that Claude Sonnet 4 did experience a dip in output quality due to a couple of bugs:

✨ Bug 1 → Affected a small % of Sonnet 4 requests starting Aug 5, got noticeably worse from Aug 29 – Sep 4.
✨ Bug 2 → Hit both Sonnet 4 and Haiku 3.5 between Aug 26 – Sep 5.

✅ Both issues have been fixed, and Anthropic clarified that quality is never intentionally degraded — this was purely a technical hiccup.

So if your Copilot completions felt “off” during that window 🤔, it wasn’t you — it really was the model. Things should now be back to normal 🚀.

[itzdheerajsingh]

Incident on July 22, 2025: GitHub experienced degraded performance specifically with the Claude Sonnet 4 model across Copilot Chat, VS Code, and other Copilot products. The issue was linked to an upstream provider but was resolved within about 16–17 minutes.
IsDown

→ If you encountered issues around that time, they were likely due to this temporary outage rather than a long-term decline in quality.

[johsephkarter7]

Yes, I’ve noticed that too outputs feel a bit less refined lately. Finding a good workaround feels like an ah bonus when it finally clicks.

[gingermiteUK]

Yes. Agent mode keeps corrupting my code when editing and having to rebuild files from scratch. Also keeps putting in StateHasChanges() (instead of StateHasChanged) and then having to correct itself. Won;t be using it again. Hope version 4.5 will be available in Co-Pilot soon and is better!