Ornith 35B Outperforms Qwen 3.6 and Laguna S in Open-Source Showdown
S_Anv · reddit · 2026-08-03
The author conducted a horizontal capability test on three open-source LLMs: Ornith 35B, Qwen 3.6 35B (A3B), and Laguna S 2.1 122B.
Tested via native llama.cpp with a single attempt, the results ranked:
- Ornith 35B (Q8 K XL quantization) performed the best.
- Qwen 3.6 35B came in second.
- Laguna S 2.1 surprisingly underperformed, with the author considering it a failure.
Live prompts and outputs are publicly available for community review.
More from Models
- Gary Marcus: OpenAI's Astra is Vastly Oversold, Math Breakthroughs Don't Mean AGI — Gary Marcus · 2026-08-03
- Only OpenAI Mastered Reasoning, Yet DeepSeek Tops Benchmark with Medium Effort — teortaxesTex · 2026-08-03
- Gemini Video Generation Backlash: AI Pro Subscribers Get Fewer Credits Than Free Users — Captain-Thump · 2026-08-03
- Fail: Google Gemini Botches a Simple Counting Task — polo421 · 2026-08-03
- Cutting AI Coding Bills from $200 to $20/Month: 105 Bugs Tested — PawelHuryn · 2026-08-03
- AI Coding for PMs: $20/Month Plan Beats $200 Setup in Bug Fixing — PawelHuryn · 2026-08-03