Qwen3.8-27B thinking xhigh Vs. thinking off - Apple M5 Max
Summary
Benchmarks on a MacBook Pro M5 Max show that disabling thinking mode in Qwen3.8-27B severely degrades output quality, while xhigh thinking mode uses 5.5x more tokens and runs 6x longer.
Similar Articles
ThinkingCap 3.8-27B vs. Swift 3.8-27B vs. Qwen 3.8-27B Benchmarks
The article benchmarks ThinkingCap-Qwen3.8-27B and Swift-Qwen3.8-27B against the original Qwen3.8-27B, showing both fine-tunes reduce reasoning tokens by ~40% with minimal performance loss, though with differences in language-specific results and token usage patterns.
Qwen3.8-27B at 144 tok/s on an M5 Max MacBook Pro
Inco Splash is an open-source inference engine optimized for Apple silicon, offering significant speed improvements for running AI models like Qwen3.8-27B on M-series MacBooks.
The difference between "medium" and "xhigh" reasoning effort for Qwen3.8-27B is actually insane.
The user observes that setting reasoning effort to 'xhigh' in Qwen3.8-27B generates significantly more thinking tokens compared to 'medium', with dramatic differences in token usage.
Qwen 3.8 27B is excellent, but it defaults to wildly overthinking things
Qwen 3.8 27B is a powerful open-source 27B parameter vision-capable LLM from Alibaba's Qwen research lab, praised for its benchmarks but criticized for defaulting to excessive reasoning effort, which slows down performance on consumer hardware.
@_akhaliq: bottlecapai/ThinkingCap-Qwen3.6-27B Capability of Qwen3.6-27B with 50% less thinking tokens on average, and over 90% le…
bottlecapai releases ThinkingCap-Qwen3.6-27B, a finetuned version of Qwen3.6-27B that achieves 50% fewer thinking tokens on average and over 90% fewer in best cases, improving efficiency.