GLM 3.8-27B quality 'absurdly superior' to 35B-A3B models, saving 22-33% tokens
JLeonsarmiento · reddit · 2026-09-12
An applied-science researcher shared a hands-on account of using GLM 3.8-27B (a Qwen-based dense model) for full research workflows, calling it 'absurdly superior' to every 3.5/3.6-35B-A3B variant (kat, Ornith/tiel, nex-2, etc.), with only Ornith coming close.
- Quality: Fully replicated 5 past projects end-to-end — workflow design, data pipelines, results analysis, report writing, and data publishing — with what the author calls a 'stupid level of attention to detail.'
- Cost: 3-4x more wall time, a real toll on daily throughput on a laptop; yet the quality gap between 3.8-27B and Z.ai's 5.3/5.3-flash is far smaller than its gap with the 35B-A3B family.
- Efficiency: At effort=medium it spends 22-33% fewer tokens with a lower RAM footprint, hitting limits and compaction less often.
The author's verdict: accept the slower speed for the quality and let the 'fat bottom Qwen' work.
More from Models
- Gemini 4 Carbon vs Opus 5.5 hands-on: 3x cheaper at $4.10 and twice as fast — vitaliychiley · 2026-10-31
- 4B model Spark-X2.6 scores 94.6/100 on community benchmark, beating 27B-35B models — mesolimbic-bliss · 2026-10-31
- opencode: $100 of usage gets you DeepSeek V4.1 Flash — intellectronica · 2026-10-31
- Researcher Finds 600+ Bugs in LLM Kernel Code; Claude Blocks Safety Analysis — pmddomingos · 2026-10-31
- Signal65: Snowflake's CoCo agent finishes workflow 4.6x faster than manual — ryanshrout · 2026-10-31
- Mistral Large 4 matches GLM on coding with 47% fewer tokens, halves preview price — petrusenko_max · 2026-10-31