Swift 1.5 matches Qwen3.8 27B quality on M5 Max while writing 34% fewer tokens
DerTomsn · reddit · 2026-09-30
A Redditor benchmarked Swift-1.5-Qwen3.8-27b (oQ8e, MLX) on an Apple M5 Max 64GB via llm-bench.io with 262k context and thinking at xhigh, against base Qwen3.8 27B.
- Token output: Swift averages 51k generated tokens per full run vs 77k for Qwen3.8 — biggest savings in code generation (28.6k vs 40.1k) and research (11.1k vs 21.2k); about 3/4 of Swift's output is still reasoning, just less of it
- Speed & quality: 34.6 vs 32.8 tok/s generation, 400 tok/s prompt processing for both, and LLM-judge scores of 85.8 vs 85.0 — a tie within run-to-run variance (84.3–87.2)
- Wall time: full benchmark averaged 24 min for Swift vs 38 min for Qwen3.8
Bottom line: equal quality, roughly a third faster thanks to leaner reasoning. The author plans to use it as a daily driver. All eight raw benchmark runs are linked.
More from Infra
- Mobile AI Factory: agents auto-build on-device SLMs with 19x less GPU time — pmttyji · 2026-11-16
- Stratechery's Fictional Obituary: Imagining Jensen Huang's 2026 Exit From Nvidia — Gregory_C_Allen · 2026-11-16
- vLLM serves DeepSeek V4.1-Flash at 954K tokens/s per GPU, $38.96/GPU-hour on GB300 NVL72 — bookwormengr · 2026-11-16
- Colossus 2's 1.2M B200s could generate ~$400B/year in tokens at DeepSeek Flash rates — bookwormengr · 2026-11-16
- BIS issues rule blocking remote access to US AI chips via third-country data centers — teortaxesTex · 2026-11-16
- Qdrant 1.20.1: 50% fewer cold-storage reads, ~2x faster quantization, agent-ready setup — qdrant_engine · 2026-11-16