Gemini 4 Neon returns finish reason [DONE] and "Sorry, no response was returned." past 200k tokens#349704
- Copilot Chat Extension Version: 0.75.0
- VS Code Version: 1.147.0 (8a14f0cc1b08d531eb83255ea3764ae22aa932b5)
- OS Version: macOS 26.7.1 (25G241), arm64
- Feature (e.g. agent/edit/ask mode): Agent mode (auto-approve)
- Selected model (e.g. GPT 4.1, Claude 3.7 Sonnet): Gemini 4 Neon (
gemini-4-neon), also seen on Gemini 4 Carbon (gemini-4-carbon)
Related to #253126.
In agent mode with Gemini 4 Neon set to the 983k context window ("Longer sessions"), requests start failing once the conversation passes ~200k tokens. The request sits for 12-300s without streaming anything, ends with finish reason: [DONE], auto-retries 3 times ("Recovered from a request error"), and then stops with Sorry, no response was returned..
Looking through my local Copilot Chat logs from Nov 11-23:
gemini-4-neonhitunknownon 81 of 1,535 completed requests (5.3%), and all 81 loggedfinish reason: [DONE].gemini-4-carbonhitunknownon 13 of 574 requests (9 withfinish reason: [DONE]).claude-opus-5.5(2,424 requests) andgpt-6.*(440 requests) in the same workspace had 0unknown/[DONE]failures.
Every failure in my session logs happened past 200k prompt tokens (219k-693k, plus one at 188k right after a 671k-char terminal output). Nothing below 200k failed. It happens at both and reasoning effort.