A Reddit user conducted an experiment grading 39 local large language models (LLMs) using five different frontier LLMs as judges. The grading process involved approximately 50 prompts, with larger models receiving more tests. The frontier LLMs, including Gemini 3.8, GPT-6.1 (Terra and Sol), and Gemini 4 (Argon), evaluated the local models on capability and foundation for fine-tuning. The user noted the cost and time involved in the grading process, with some frontier LLMs having token limitations that affected their participation. AI
IMPACT Provides insights into the relative performance and suitability for fine-tuning of various local LLMs compared to frontier models.
RANK_REASON User-conducted benchmark and evaluation of multiple LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
- Argon (Gemini 4)
- Bonsai-2 27B PQ2
- Claude
- Gemini 3.8
- Gemma-4 12B Q6
- Nanbeige4.2 3B
- Qwen3.6 35B-A3B
- Sol (GPT-6.1 Sol)
- Terra (GPT-6.1)
- Underdog-Saluki 27B
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
