Nathan Esquenazi 273819a975 gh-pages: fix local models section accuracy ...

- Qwen 3.5 27B: context 800K→262K native, add 85.5% GPQA score, clarify 24GB GPU needed
- DeepSeek R1 32B distill: VRAM 17→20 GB, note 82M+ Ollama pulls, add Ollama run command
- Llama 4 Scout: clarify Ollama serves 128K (not 10M), add 67GB download size note
- Phi-4 Reasoning: call out 'plus' variant (phi4-reasoning:14b-plus), add 81.3% AIME / 68.9% GPQA scores
- Fix 800K context pill on Qwen card

2026-04-14 01:39:15 +00:00

..

index.html

gh-pages: fix local models section accuracy

2026-04-14 01:39:15 +00:00