Engineer / developer observations of Gemma4-31B, Qwen3.8-27B, and 6.1-Sol for software engineering work
Models: Gemma4-31B vs Qwen3.8-27B at the same quantization (an Unsloth flavor of Q4).
Proof1 community thread
Key points
- Today I decided to do a few tests on things that I'd consider analogous to "real" work that I do, trying out some different combinations of models and harnesses.
- Harnesses: Codex CLI vs OpenCode CLI
- Types of Work: "Tell me about this code" and "Let's make something new"
- I thought of an application that's concisely scoped, doesn't rely on a legacy codebase or significant external dependencies, and would probably take me ~1-2 days to hammer out by hand. ~8 paragraphs of prompt covering general concept, user experience in a couple different roles, design constraints and future-proofing needs.
Sources (1)
- [1]Engineer / developer observations of Gemma4-31B, Qwen3.8-27B, and 6.1-Sol for software engineering workr/LocalLLaMA (top, daily) · Oct 10, 06:49 PM
Models: Gemma4-31B vs Qwen3.8-27B at the same quantization (an Unsloth flavor of Q4).
Today I decided to do a few tests on things that I'd consider analogous to "real" work that I do, trying out some different combinations of models and harnesses.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 10, 2026vllm-sr/d3
- Oct 9, 2026Microsoft's Decision-1 model enters the fast-growing AI decision model race
- Oct 9, 2026A new feature for my blog, built using my voice
- Oct 9, 2026Taking Tokenization off the Host for Agentic LLM Serving
- Oct 9, 2026Qwen/Qwen-Image-2.1-Turbo
- Oct 8, 2026[AINews] not much happened today
