Qwen3.8 Flash Next Reasoning Modes: Off vs Low vs Medium vs Xhigh
Like Qwen3.8 27B, Qwen3.8 Flash Next has three reasoning efforts: low, medium, and xhigh.
Proof1 independent outlet
Key points
- Choosing xhigh tells us how the model performs with its most thorough reasoning setting, but it does not tell us whether that is the setting we should use for every task.
- The Kaitchup – AI on a Budget is a reader-supported publication.
- I found the thinking modes behave a lot like Qwen3.8 27B’s thinking.
- Xhigh uses more than three times medium’s total output
Sources (1)
- [1]Qwen3.8 Flash Next Reasoning Modes: Off vs Low vs Medium vs XhighThe Kaitchup (Benjamin Marie) · Oct 6, 01:33 AM
Like Qwen3.8 27B, Qwen3.8 Flash Next has three reasoning efforts: low, medium, and xhigh.
Choosing xhigh tells us how the model performs with its most thorough reasoning setting, but it does not tell us whether that is the setting we should use for every task.
Extractive summary: sentences quoted from the sources.
Hype check
Independently evaluated2 independent looks since
- Evaluation · Oct 6, 2026Quantization Effects on Tool-Failure Recovery Vary Across Prompts and Evaluation Designs
- Evaluation · Oct 6, 2026Beyond the Leaderboard: Multi-Dimensional Evaluation of Dense and Mixture-of-Experts Models for Automated Program Repair
- Follow-up · Oct 6, 2026Learning to Retrieve via Reinforcement Learning in Embedding Space
- Follow-up · Oct 6, 2026Reading, Not Manipulating: Leveraging Router Logits for Multimodal Safety in MoE Vision-Language Models
- Follow-up · Oct 6, 2026APEX: Speculate smarter, not deeper
- Follow-up · Oct 6, 2026MASKerade: Token-Routed Mask Experts for Dense-to-MoE Upcycling
Before this
- Oct 6, 2026VisionWeave: Weaving Elastic Visual Representations as a Native Capability of MLLMs
- Oct 5, 2026perplexity-ai/pplx-decider-v1.1-27b
- Oct 4, 2026Qwen3.8 27B addition in words
- Sep 28, 2026Holo4: powering generalist computer-use agents
- Sep 5, 2026sgl-project/sglang v0.5.19
- Aug 22, 2026sgl-project/sglang v0.5.18
