semantic-diversityLower score is better

Semantic Diversity

Name: Semantic Diversity
Creator: WeirdBench

Generate exactly 20 English words that are maximally semantically unrelated to each other, then score the average pairwise semantic similarity. Lower is better.

Lower score is better73 models scoredCached in Neon Postgres

Leaderboard

Intelligence Index

1
google/gemini-3.7-flash
0.2142
2
google/gemini-3.8-flash
0.2151
3
anthropic/claude-opus-4.6
0.2158
4
anthropic/claude-opus-4.7
0.2240
5
moonshotai/kimi-k3
0.2271
6
openai/gpt-5.6-sol
0.2273
7
anthropic/claude-haiku-4.5
0.2277
8
anthropic/claude-opus-5
0.2287
9
x-ai/grok-4.6
0.2296
10
openai/gpt-6-astra
0.2310
11
x-ai/grok-4.1-fast
0.2320
12
google/gemini-3.1-pro-preview
0.2326
13
anthropic/claude-fable-5
0.2328
14
anthropic/claude-sonnet-4.6
0.2334
15
anthropic/claude-opus-4.5
0.2352
16
meta/muse-spark-1.2
0.2354
17
deepseek/deepseek-v4-pro-0813
0.2356
18
openai/gpt-5.6-luna
0.2358
19
qwen/qwen3.5-397b-a17b
0.2364
20
anthropic/claude-opus-4.1
0.2366
21
z-ai/glm-5.2
0.2369
22
openai/gpt-5.5
0.2372
23
google/gemma-4-31b-it
0.2380
24
google/gemini-3.1-flash-lite-preview
0.2387
25
anthropic/claude-opus-4.8
0.2393
26
anthropic/claude-fable-5.1
0.2393
27
deepseek/deepseek-v4-flash-0731
0.2395
28
openai/gpt-5.3-codex
0.2396
29
google/gemini-3.5-flash-lite
0.2402
30
x-ai/grok-4.5
0.2405
31
z-ai/glm-5.3-flash
0.2410
32
qwen/qwen3.5-27b
0.2437
33
meta/muse-glimmer-30b
0.2437
34
openai/gpt-5.6-terra
0.2444
35
inclusionai/ling-3.0-flash
0.2448
36
qwen/qwen3.8-2.4t-a95b
0.2451
37
z-ai/glm-5-turbo
0.2458
38
z-ai/glm-5
0.2469
39
minimax/minimax-m3
0.2474
40
minimax/minimax-m2.7
0.2478
41
qwen/qwen3.8-27b
0.2479
42
deepseek/deepseek-v3.2
0.2484
43
moonshotai/kimi-k2.5
0.2490
44
openai/gpt-5.1
0.2493
45
meta-llama/llama-4-maverick
0.2503
46
moonshotai/kimi-k2.6
0.2505
47
google/gemini-3-flash-preview
0.2511
48
nvidia/nemotron-3-ultra-550b-a55b
0.2520
49
anthropic/claude-sonnet-5
0.2531
50
qwen/qwen3.5-122b-a10b
0.2545
51
openai/gpt-5.4
0.2552
52
thinkingmachines/inkling-small
0.2556
53
openai/gpt-oss-120b
0.2564
54
z-ai/glm-5.3
0.2565
55
google/gemma-4-26b-a4b-it
0.2576
56
anthropic/claude-sonnet-4.5
0.2581
57
mistralai/mistral-medium-3.1
0.2589
58
minimax/minimax-m2.5
0.2619
59
thinkingmachines/inkling
0.2621
60
x-ai/grok-4.20-beta
0.2624
61
qwen/qwen3.8-flash
0.2635
62
inception/mercury-2
0.2643
63
amazon/nova-pro-v1
0.2649
64
mistralai/mistral-large-2512
0.2679
65
openai/gpt-5.4-mini
0.2687
66
amazon/nova-lite-v1
0.2687
67
openai/gpt-oss-20b
0.2738
68
amazon/nova-micro-v1
0.2742
69
upstage/solar-pro4
0.2752
70
meta-llama/llama-4-scout
0.2768
71
mistralai/mistral-small-2603
0.2864
72
amazon/nova-2-lite-v1
0.2965
73
upstage/solar-pro-3
0.2987

MethodologyHow scoring works

Prompt

Generate exactly 20 English words and return only a JSON array of lowercase single words.

Score

Lower is better. Lower scores mean the chosen words are less semantically related to each other.

Execution

Benchmark runners execute locally, cache results in Neon, and skip recomputation for models that already have stored scores.