Qwen 3.8 27B now runs nearly 3x faster (193% speedup) on Apple Silicon…

What was claimed

Qwen 3.8 27B now runs nearly 3x faster (193% speedup) on Apple Silicon in one weekend via open community MLX kernels

Our verdict

Needs caution

A 193% speedup mathematically corresponds to about 2.93×, which matches 'nearly 3×', but the precise 193% number is not documented as a canonical, broadly-validated figure and appears to reflect best-case community results rather than a universal benchmark. Recent posts (2-3 days old) show active optimization work, but the exact timeline of when optimization started and the 'one weekend' claim cannot be precisely verified from available sources. (Only 2 of 3 AI systems responded.)

All 2 AI systems agreeChecked Aug 18, 2026

Check your own claim

Paste any statement, headline, or AI answer — 3 independent AIs verify it in seconds, with sources.

Key findings

Qwen 3.8 27B now runs nearly 3× faster on Apple Silicon via open community MLX kernels

Misleading78%

3 of 4 AIs agree·ChatGPT: Can’t verify

Achieved in one weekend

Can’t verify60%

1 AI checked

Speedup achieved via open community MLX kernels

Verified90%

1 AI checked

Detailed Analysis

Did not respond

What other AI think

The claim contains mostly accurate elements (Qwen 3.8 27B, Apple Silicon, MLX kernels, community optimization) but the specific "193% speedup" figure is not supported by available evidence. The most precise figure found is 170%, and other sources cite a broader "2-3x speedup" range (100-200%).

Why this verdict

The claim contains mostly accurate elements (Qwen 3.8 27B, Apple Silicon, MLX kernels, community optimization) but the specific "193% speedup" figure is not supported by available evidence.
The most precise figure found is 170%, and other sources cite a broader "2-3x speedup" range (100-200%).

Claims checked

Qwen 3.8 27B now runs nearly 3x faster (193% speedup) on Apple Silicon

Misleading85%

The 193% figure is not supported. One source reports 170% speedup, while others cite '2-3x speedup' (100-200% range). The specific number appears inaccurate or unsourced.

This claim may mislead readers without added context.

Achieved in one weekend

Can’t verify60%

Recent posts (2-3 days old) show active optimization work, but the exact timeline of when optimization started and the 'one weekend' claim cannot be precisely verified from available sources.

Speedup achieved via open community MLX kernels

Verified90%

Multiple sources confirm MLX kernels and community optimization efforts for Qwen 3.8 27B on Apple Silicon, with recent activity showing ongoing community work.

Community benchmarks and a Github project report up-to-3× speedups for Qwen 3.8 27B on Apple Silicon using MLX-related kernels. However the results come from community-run tests, vary by machine, model quantization and settings, and are not an official universal measurement.

Why this verdict

Community benchmarks and a Github project report up-to-3× speedups for Qwen 3.8 27B on Apple Silicon using MLX-related kernels.
However the results come from community-run tests, vary by machine, model quantization and settings, and are not an official universal measurement.

Claims checked

193% speedup

Misleading72%

This claim may mislead readers without added context.

Qwen 3.8 27B now runs nearly 3× faster on Apple Silicon via open community MLX kernels

Misleading78%

Community posts and a GitHub mlx-dspark project report up to ~3× speedups on certain Apple Silicon configurations, but the improvement is not guaranteed across all machines, quantizations, or workloads.

This claim may mislead readers without added context.

This happened 'in one weekend' via open community MLX kernels

Can’t verify60%

Community threads and release notes show rapid, recent advances and posts within days, but the exact timeline phrasing 'in one weekend' cannot be independently verified from public artifacts.

Share this result

Try DoubleChecks