Qwen3.8-27B vs Frontier vs Others
Here're results from my final test of coding, coding (panda/numpy), data analysis, data-driven decision-making, and local RAG domains I interact with most often.
Proof1 community thread
Key points
- When Qwen4 or others get released, if they do better on this, gives me real good feel for my use case.
- Models on OpenRouter sometimes disclose their quantization and sometimes not. - or-qwen-qwen3.8-27b is stock at bf16
- iq3 GSQ-RCO is decent performer even placed among Q4s
- Muse Glimmer, Qwen3.6-35B-A3B etc. included for comparison.
Sources (1)
- [1]Qwen3.8-27B vs Frontier vs Othersr/LocalLLaMA (top, daily) · Oct 12, 06:26 PM
Here're results from my final test of coding, coding (panda/numpy), data analysis, data-driven decision-making, and local RAG domains I interact with most often.
When Qwen4 or others get released, if they do better on this, gives me real good feel for my use case.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 12, 2026openbmb/MiniCPM-V-4.7-35B-A3B
- Oct 12, 2026[AINews] Meta Muse Wonder & Wonder Full: MSL’s Claude 5.5-tier Frontier Model
- Oct 11, 2026vllm-ascend updates (GLM5.3-flash) on dual 310p Ascend cards
- Oct 11, 2026UPDATE: Qwen 3.8 27B 140 tok/s on single RTX 3090 Megakernel: KL divergence 0.0009 vs llama.cpp
- Oct 11, 2026Running Next Flash IQ3_XXS at ~70 tok/s with 100k context or 2 instances of Qwen 3.6 35B A3B IQ4 at ~145 tok/s with 256k all on $500 of ex mining BC-250 boards
- Oct 10, 2026vllm-sr/d3
