Qwen 3.8 27B is missing a 'high' reasoning effort between medium and xhigh
alanoo · reddit · 2026-08-19
A Redditor asks why Qwen 3.8 27B has no 'high' reasoning effort option. The gap between 'medium' (which barely thinks) and the default 'xhigh' (already the subject of many complaints about overthinking) is described as ridiculously huge—and the naming itself implies a missing 'high' mode.
More from Models
- Grok Bot Ran a 12-Hour Data Analysis Nonstop, Handling API Rate Limits Itself — Daniel_Farinax · 2026-10-16
- One year of Recursive Language Models: GPT-5-mini beats GPT-5 by 110% on 132k-token tasks, cheaper too — a1zhang · 2026-10-16
- Users Report Claude Models Suddenly More Compliant After Forced Compaction — RealSharpNinja · 2026-10-16
- Prediction: Opus 5.5-level models on laptops by 2027, backed by Qwen 27B data — cephaloform · 2026-10-16
- Users Say Krea 2 Output Degraded Sharply Since Krea 3 Beta Launched — szywis · 2026-10-16
- LLM-as-a-Verifier lifts Nemotron 3.5 Super by 9-10 points on agentic benchmarks — simonguozirui · 2026-10-16