nutrition-predictionHigher score is better

Nutrition Prediction

Name: Nutrition Prediction
Creator: WeirdBench

Predict calories, protein, carbs, and fat from ingredient lists for a fixed 50-dish Nutrition5k sample. Higher is better.

Higher score is better70 models scoredCached in Neon Postgres

Leaderboard

Intelligence Index

1
qwen/qwen3.8-27b
25.7670
2
anthropic/claude-opus-5
23.8534
3
mistralai/mistral-small-2603
23.8232
4
openai/gpt-6-astra
23.7566
5
openai/gpt-oss-20b
23.7467
6
openai/gpt-5.6-sol
23.6810
7
z-ai/glm-5-turbo
23.5975
8
qwen/qwen3.5-122b-a10b
23.4188
9
x-ai/grok-4.6
23.0717
10
mistralai/mistral-medium-3.1
22.9013
11
google/gemini-3.8-flash
22.8262
12
x-ai/grok-4.20-beta
22.5383
13
thinkingmachines/inkling-small
22.2876
14
anthropic/claude-opus-4.8
22.2521
15
meta-llama/llama-4-maverick
22.0002
16
inception/mercury-2
21.8978
17
x-ai/grok-4.5
21.8124
18
amazon/nova-lite-v1
21.6084
19
openai/gpt-5.4
21.5840
20
anthropic/claude-opus-4.1
21.5560
21
anthropic/claude-opus-4.7
21.5195
22
mistralai/mistral-large-2512
21.4432
23
openai/gpt-oss-120b
21.3529
24
openai/gpt-5.5
21.2108
25
qwen/qwen3.8-2.4t-a95b
21.2042
26
amazon/nova-pro-v1
21.1849
27
openai/gpt-5.3-codex
21.1440
28
moonshotai/kimi-k3
21.1347
29
upstage/solar-pro4
21.1071
30
google/gemini-3.7-flash
21.0439
31
anthropic/claude-fable-5
20.9741
32
deepseek/deepseek-v4-pro-0813
20.9608
33
deepseek/deepseek-v4-flash-0731
20.8675
34
anthropic/claude-sonnet-4.6
20.7120
35
anthropic/claude-haiku-4.5
20.7097
36
google/gemini-3.1-flash-lite-preview
20.5872
37
google/gemini-3.5-flash-lite
20.5695
38
moonshotai/kimi-k2.5
20.1413
39
google/gemini-3-flash-preview
19.9842
40
openai/gpt-5.1
19.9581
41
google/gemma-4-26b-a4b-it
19.8426
42
qwen/qwen3.8-flash
19.7320
43
z-ai/glm-5.3
19.6902
44
google/gemma-4-31b-it
19.6220
45
anthropic/claude-opus-4.6
19.5945
46
z-ai/glm-5
19.4215
47
anthropic/claude-sonnet-4.5
19.3172
48
upstage/solar-pro-3
19.1532
49
moonshotai/kimi-k2.6
19.1032
50
openai/gpt-5.6-luna
19.0885
51
anthropic/claude-opus-4.5
18.8760
52
z-ai/glm-5.3-flash
18.7485
53
anthropic/claude-sonnet-5
18.6307
54
anthropic/claude-fable-5.1
18.5766
55
qwen/qwen3.5-397b-a17b
18.5588
56
minimax/minimax-m3
18.4202
57
x-ai/grok-4.1-fast
18.3064
58
inclusionai/ling-3.0-flash
18.2626
59
minimax/minimax-m2.7
17.5755
60
deepseek/deepseek-v3.2
17.4072
61
google/gemini-3.1-pro-preview
17.1302
62
z-ai/glm-5.2
16.8840
63
meta/muse-spark-1.2
16.7887
64
meta-llama/llama-4-scout
16.6109
65
amazon/nova-2-lite-v1
16.0830
66
openai/gpt-5.4-mini
15.2246
67
amazon/nova-micro-v1
14.8388
68
minimax/minimax-m2.5
14.6662
69
openai/gpt-5.6-terra
13.6362
70
meta/muse-glimmer-30b
-2.5157

MethodologyHow scoring works

Prompt

Given only the ingredient list, return JSON with numeric `calories`, `protein`, `carbs`, and `fat` fields and no extra text.

Score

Higher is better. Overall score is 60% accuracy and 40% average correlation, where accuracy = 100 / (1 + average MAPE percentage).

Execution

Benchmark runners execute locally, use OpenRouter for predictions, fetch the fixed Nutrition5k metadata sample, cache results in Neon, and skip recomputation for models that already have stored scores.