Reddit user finds popular Qwen 27B finetune underperforms stock after thinking time cut
Brief-Effect9065 · reddit · 2026-09-03
A Reddit user reports that DavidAU's Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored finetune is a disappointment, suspecting that drastically shortening the model's thinking duration effectively lobotomized it.
Testing single-file HTML tower defense game generation in the DeepSeek harness — with both simple and detailed prompts, in low and xhigh modes — the finetune performed worse than both the stock model and its predecessor (Qwen3.8-27B-Cold-Fusion-GAIN-V1.1). The poster questions where the model's published benchmark scores come from.
More from Models
- Rival AI agents: cross-vendor model review catches what self-review misses — rseroter · 2026-09-03
- Gemini's Distinctive Take on AI Sentience Turns Heads — aiamblichus · 2026-09-03
- Grok Heavy user burns through limits in 3-4 days, suspects a metering bug — Daniel_Farinax · 2026-09-03
- Reddit users report ChatGPT 5.6 suddenly got much worse: forgetful, lazy and hallucinating — Individual-Value7169 · 2026-09-03
- Claude Fable 5.1 tops Senior SWE-Bench, matching Fable 5 at ~50% cost via Medium effort tier — ajratner · 2026-09-03
- Fable 5.1 called Anthropic's best vision model yet, with big gains in detection and counting — airesearch12 · 2026-09-03