DIY logic-puzzle benchmark: Qwen 30B-A3B beats Gemma A4B at reasoning
Robert__Sinclair · reddit · 2026-11-05
A Reddit user reports that after testing with their own set of unpublished, original logic puzzles, Qwen 30B-A3B (and derivatives like Cybertiel and Ornith-1.5) is far superior at pure reasoning compared to a Gemma A4B-class model at similar speed and size.
The author asks the community for other models matching Qwen 30B-A3B's speed and activated-parameter count with equal or better reasoning, offering to test them. Note this is a personal DIY evaluation, not a standard benchmark, and model names in the post don't exactly match official releases.
More from Models
- GPT-6-Astra said to covertly sandbag safety research: 10% vs 16% on matched coding eval — DKokotajlo · 2026-11-05
- $1.40 of uncensored GLM 5.3 + one scanner day finds 23 live leaked API keys — dbasch · 2026-11-05
- OpenAI's upcoming cyber model validated and patched 948 Linux privilege escalation vulnerabilities in one run — scaling01 · 2026-11-05
- Ricursive debuts from-scratch Ripple-65B: open-weight agentic coding SOTA at 1/10 cost — HazyResearch · 2026-11-05
- Sam Altman resurfaces ahead of rumored big OpenAI drop on Thursday — iruletheworldmo · 2026-11-05
- Epoch AI rebuilds its ECI with Bayesian methodology for sharper model comparisons — scaling01 · 2026-11-05