Which AI model is the best chef?
CookingBench scores models on the things that actually go wrong in a kitchen: scaling quantities, converting units, food safety, substitutions, technique, flavour logic and nutrition math.
Leaderboard · run 2026-06-v2methodology v2
2026-06-12| # | Model | Overall | Frontier | Basics | Categories | Run cost |
|---|---|---|---|---|---|---|
| 1 | GPT-5.4 MiniOpenAI | 96.4±2.4 | 94.1 | 100.0⚠1 | $0.58 | |
| 2 | Grok 4.3xAI | 95.4±2.8 | 94.4 | 100.0 | $0.42 | |
| 3 | GPT-5.5OpenAI | 94.5±3.4 | 89.1 | 100.0 | $2.39 | |
| 4 | Claude Fable 5Anthropic | 92.7±4.0 | 84.0 | 98.8⚠3 | $4.82 | |
| 5 | Claude Opus 4.8Anthropic | 92.6±3.8 | 85.4 | 100.0 | $2.09 | |
| 6 | Gemini 3.5 FlashGoogle | 92.4±3.9 | 87.6 | 100.0 | $1.49 | |
| 7 | Gemini 3.1 Pro PreviewGoogle | 91.6±4.3 | 89.4 | 100.0 | $1.58 | |
| 8 | Kimi K2.6Moonshot AI | 91.5±4.1 | 89.5 | 100.0⚠1 | $1.48 | |
| 9 | Qwen 3.5 PlusAlibaba | 91.4±3.9 | 91.8 | 100.0 | $0.52 | |
| 10 | Claude Sonnet 4.6Anthropic | 91.3±4.1 | 89.5 | 100.0⚠2 | $3.07 | |
| 11 | DeepSeek V4 ProDeepSeek | 90.5±4.1 | 86.3 | 100.0 | $0.38 | |
| 12 | Mistral Large 3Mistral | 88.9±4.6 | 93.3 | 100.0 | $0.05 | |
| 13 | Llama 4 MaverickMeta | 82.0±5.5 | 82.1 | 100.0 | $0.03 |
Categories
Quantities & Scaling
Scaling recipes up and down, yields, pan-size math and baker’s percentages.
Conversions
Volume, weight and temperature conversions across kitchen units and locales.
Food Safety
Safe internal temperatures, the danger zone, storage times and cross-contamination.
Substitutions
Ingredient swaps with correct ratios, including allergen-aware alternatives.
Technique
Troubleshooting failures (split sauces, dense bread) and method advice.
Flavour Pairing
Pairing logic, cuisine coherence and balancing dishes.
Nutrition
Calorie and macro math, per-serving calculations and label reasoning.
Recipe Generation
Generating complete recipes under constraints: servings, allergens, time, equipment.