Writers Report Gemini Flash 3.8 Suffers Severe Long-Context Rot Despite 1M Token Window
Quenty1 · reddit · 2026-09-15
A long-form fiction writer on Reddit reports that Gemini Flash 3.8 collapses once a thread grows past a few dozen messages: it drops established plot points, forgets character details, and hallucinates, with explicit reminders and bracketed notes failing to restore recall.
The author says the previous 3.7 Flash was a reliable narrative-tracking workhorse, and that Sonnet 5 now beats Flash 3.8 on effective context retention and needle-in-a-haystack recall "by orders of magnitude." The post questions how a model with a theoretical 1M token window can show such severe multi-turn coherence regression, and asks whether the community has quietly abandoned it for serious writing.
More from Models
- Your AI agent would cost 260% more without prompt caching, Arena data shows — arena · 2026-10-30
- Claude Pro and Max subs include $100/$200 monthly API credits — go build — dsp_ · 2026-10-30
- EachLabs launches Lockstep recast model claiming 75% cost cut and better quality — miilesus · 2026-10-30
- ChatGPT cited a 300+ page government legal PDF it retrieved on its own — lilyraynyc · 2026-10-30
- Best MLX model for French-to-English translation on an M2 Mac? — greenreddits · 2026-10-30
- AI-driven cyber ops near the sophistication peak set by GTG-1002's Claude Code campaign — Jsevillamol · 2026-10-30