CookingBench

Archived v2.1 result / categories

Technique

Troubleshooting failures (split sauces, dense bread) and method advice.

Original v2.1 category means (archived, not a ranking)

Sorted by score for readability. Category means were never tested for statistical separation; over as few as a dozen questions, an ordering at this granularity would be noise presented as precision.

Question heatmap (public questions only)

Model001002003004005006007008009010011012013014015016
Grok 4.5
GPT-5.6 Terra Pro
Claude Opus 5
GPT-5.6 Sol Pro
Kimi K3
Gemini 3.6 Flash
Claude Fable 5
Qwen 3.7 Max
GPT-5.4 Mini
Gemini 3.1 Pro Preview
Claude Sonnet 5
DeepSeek V4 Pro
Mistral Large 3
Llama 4 Maverick

Each cell is one question; deeper colour = higher score. Hover for exact values.