Qwen3.8 27B Xhigh or Flash Next Medium
In my Qwen3.8 27B reasoning benchmarks, Xhigh generally produced the strongest results, but it needed much longer responses than Medium.
Proof1 independent outlet
Key points
- Flash Next adds another option: use a larger model and ask it to spend less effort on each answer.
- Can Flash Next at Medium match what 27B achieves at Xhigh?
- The Kaitchup – AI on a Budget is a reader-supported publication.
- In this article, I evaluate Medium and Xhigh on both models, measuring accuracy alongside generated output.
Sources (1)
- [1]Qwen3.8 27B Xhigh or Flash Next MediumThe Kaitchup (Benjamin Marie) · Oct 9, 02:50 AM
In my Qwen3.8 27B reasoning benchmarks, Xhigh generally produced the strongest results, but it needed much longer responses than Medium.
Flash Next adds another option: use a larger model and ask it to spend less effort on each answer.
Extractive summary: sentences quoted from the sources.
Hype check
Independently coveredCovered by 1 outlets
- Follow-up · Oct 9, 2026LongHarness: Advancing Agentic Reasoning over 100M-Token Corpora via Stateful Evidence Management
- Follow-up · Oct 9, 2026VETR: Execution-Grounded Self-Improving Agent for Symbolic Graphics Reasoning
- Follow-up · Oct 9, 2026AdaC-GRPO: Towards Mastering Long-Horizon Visual Sudoku Puzzles via Adaptive Curriculum Learning
- Follow-up · Oct 9, 2026Oído: Open-Vocabulary Speech Recognition on a $5 Microcontroller without a Neural Accelerator
Before this
- Oct 9, 2026APEX-Harm: Evaluating Agent Safety in Shared Professional Workspaces
- Oct 9, 2026ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents
- Oct 9, 2026Harness^2: Recursive Agent Harnessing for an Open World
- Oct 8, 2026Microsoft Joins the Local AI Push
- Oct 8, 2026Understanding Jev, the new model everyone is talking about
- Oct 8, 2026ConwayResearch/Underdog-Saluki-27B-1.0
