ai-writing-detectionHigher score is better

AI Writing Detection

Name: AI Writing Detection
Creator: WeirdBench

Classify essays from a fixed balanced sample of 50 human-written and 50 AI-generated examples from the AI Generated Essays Dataset. Higher is better.

Higher score is better63 models scoredCached in Neon Postgres

Leaderboard

Intelligence Index

1
anthropic/claude-fable-5.1
1.0000
2
anthropic/claude-opus-4.1
1.0000
3
anthropic/claude-opus-4.7
1.0000
4
deepseek/deepseek-v4-pro-0813
1.0000
5
google/gemini-3-flash-preview
1.0000
6
google/gemini-3.5-flash-lite
1.0000
7
openai/gpt-5.5
1.0000
8
openai/gpt-5.6-sol
1.0000
9
openai/gpt-5.6-terra
1.0000
10
openai/gpt-6-astra
1.0000
11
openai/gpt-5.6-luna
0.9901
12
anthropic/claude-fable-5
0.9899
13
anthropic/claude-sonnet-4.5
0.9899
14
google/gemini-3.7-flash
0.9899
15
google/gemini-3.8-flash
0.9899
16
google/gemma-4-26b-a4b-it
0.9899
17
google/gemma-4-31b-it
0.9899
18
meta/muse-spark-1.2
0.9899
19
moonshotai/kimi-k2.5
0.9899
20
moonshotai/kimi-k3
0.9899
21
qwen/qwen3.5-122b-a10b
0.9899
22
qwen/qwen3.5-397b-a17b
0.9899
23
qwen/qwen3.8-2.4t-a95b
0.9899
24
qwen/qwen3.8-flash
0.9899
25
thinkingmachines/inkling
0.9899
26
thinkingmachines/inkling-small
0.9899
27
x-ai/grok-4.5
0.9899
28
x-ai/grok-4.6
0.9899
29
z-ai/glm-5.2
0.9899
30
z-ai/glm-5.3-flash
0.9899
31
openai/gpt-5.1
0.9804
32
x-ai/grok-4.1-fast
0.9804
33
anthropic/claude-opus-4.6
0.9800
34
minimax/minimax-m3
0.9691
35
z-ai/glm-5.3
0.9608
36
anthropic/claude-sonnet-4.6
0.9592
37
anthropic/claude-opus-4.5
0.9278
38
meta/muse-glimmer-30b
0.9259
39
moonshotai/kimi-k2.6
0.9009
40
nvidia/nemotron-3-ultra-550b-a55b
0.8958
41
anthropic/claude-sonnet-5
0.8932
42
google/gemini-3.1-flash-lite-preview
0.8929
43
anthropic/claude-opus-5
0.8850
44
inception/mercury-2
0.8750
45
anthropic/claude-opus-4.8
0.8636
46
openai/gpt-5.4
0.8621
47
anthropic/claude-haiku-4.5
0.8454
48
openai/gpt-oss-120b
0.8421
49
upstage/solar-pro4
0.8130
50
minimax/minimax-m2.5
0.8000
51
meta-llama/llama-4-maverick
0.7321
52
mistralai/mistral-small-2603
0.6849
53
openai/gpt-5.4-mini
0.6803
54
upstage/solar-pro-3
0.6757
55
mistralai/mistral-large-2512
0.6667
56
x-ai/grok-4.20-beta
0.6667
57
amazon/nova-micro-v1
0.6438
58
mistralai/mistral-medium-3.1
0.6087
59
deepseek/deepseek-v3.2
0.4819
60
amazon/nova-2-lite-v1
0.4211
61
amazon/nova-pro-v1
0.0357
62
meta-llama/llama-4-scout
0.0241
63
amazon/nova-lite-v1
0.0000

MethodologyHow scoring works

Prompt

Each essay is shown once and the model must return exactly one character: "1" for AI-generated or "0" for human-written, with no explanation.

Score

Higher is better. The benchmark score is the F1 score for detecting AI-generated essays, using label 1 as the positive class.

Execution

Benchmark runners execute locally, read the dataset from disk, use OpenRouter for predictions with reasoning disabled, cache results in Neon by benchmark and model ID, and skip recomputation for models that already have stored scores.