Mystery: examining a crime scene

Bad Writing Habits

Detects common prose quality anti-patterns in AI-generated creative writing, including passive voice, past progressive overuse, weak dialogue tags, filter words, purple prose, cliches, AI-ism words/adverbs/names, and more.

0-shot Creative writingRule following

Performance Score Distribution (Top 20)

Click a model name to view its detail page.

	Score
GPT-5.1	90%
Cohere Command R+ (Aug. 2024)	89%
GPT-5	88%
Claude Sonnet 4.6	85%
Qwen 3.5 397B A17B	85%
Mistral Medium 3.1	84%
Claude Opus 4.6	85%
Writer: Palmyra X5	83%
Mistral Large 2	84%
o4 Mini	84%
Hermes 3 405B	82%
MoonshotAI: Kimi K2.5	82%
DeepSeek V3 (2025-03-24)	82%
Claude Haiku 4.5	80%
Mistral Large	84%
Mistral Small Creative	82%
GPT-4.1	80%
Mistral Large 3	81%
Z.AI GLM 5	83%
GPT-5.2	82%

	Score	Cost	Time
Cohere Command R+ (Aug. 2024)	89%	$0.017	52.5s
Mistral Medium 3.1	84%	$0.0039	28.3s
Mistral Small Creative	82%	$0.0005	6.9s
Hermes 3 405B	82%	$0.0019	37.5s
Writer: Palmyra X5	83%	$0.010	20.7s
Mistral Large 3	81%	$0.0024	24.5s
Mistral Large 2	84%	$0.0090	23.4s
DeepSeek V3 (2025-03-24)	82%	$0.0009	50.3s
ByteDance Seed 1.6 Flash	83%	$0.0011	23.5s
Claude Haiku 4.5	80%	$0.0091	19.3s
Mistral Large	84%	$0.0094	24.0s
Gemini 2.5 Flash	79%	$0.0051	11.5s
o4 Mini	84%	$0.014	24.5s
Claude Sonnet 4.6	85%	$0.026	39.4s
Rocinante 12B	82%	$0.0008	19.4s
Claude 3 Haiku	79%	$0.0022	19.2s
Z.AI GLM 5	83%	$0.0068	1.0m
Gemma 3 27B	81%	$0.0004	52.4s
Hermes 3 70B	78%	$0.0006	34.2s
Ministral 3 8B	79%	$0.0003	5.1s

	Score	Cost	Speed	Stability
Cohere Command R+ (Aug. 2024)	89%	$0.017	52.5s	81%
Mistral Large 2	84%	$0.0090	23.4s	83%
Mistral Medium 3.1	84%	$0.0039	28.3s	81%
GPT-5.1	90%	$0.053	2.2m	87%
o4 Mini	84%	$0.014	24.5s	78%
Mistral Small Creative	82%	$0.0005	6.9s	77%
Mistral Large	84%	$0.0094	24.0s	77%
Claude Sonnet 4.6	85%	$0.026	39.4s	79%
Mistral Large 3	81%	$0.0024	24.5s	78%
Z.AI GLM 4.6	82%	$0.0056	1.0m	78%
Z.AI GLM 5	83%	$0.0068	1.0m	77%
ByteDance Seed 1.6 Flash	83%	$0.0011	23.5s	74%
DeepSeek V3 (2025-03-24)	82%	$0.0009	50.3s	76%
Hermes 3 405B	82%	$0.0019	37.5s	75%
Gemma 3 27B	81%	$0.0004	52.4s	77%
Grok 4,1 Fast	81%	$0.0016	34.8s	76%
Writer: Palmyra X5	83%	$0.010	20.7s	73%
Claude 3 Haiku	79%	$0.0022	19.2s	76%
MoonshotAI: Kimi K2.5	82%	$0.014	1.1m	77%
GPT-5 Mini	81%	$0.010	1.3m	79%

Rank	Model	Avg. Cost	Avg. Time	Stability	# 1	# 2	# 3	# 4	# 5	Total
4	GPT-5.1	$0.053	2.2m	87%	93	92	92	89	85	90%
1	Cohere Command R+ (Aug. 2024)	$0.017	52.5s	81%	94	92	88	86	83	89%
24	GPT-5	$0.067	2.7m	83%	93	89	88	87	84	88%
30	Qwen 3.5 397B A17B	$0.011	5.2m	81%	89	86	86	83	82	85%
37	Claude Opus 4.6	$0.073	1.2m	80%	90	86	85	83	82	85%
8	Claude Sonnet 4.6	$0.026	39.4s	79%	90	88	87	83	78	85%
5	o4 Mini	$0.014	24.5s	78%	90	84	84	81	81	84%
7	Mistral Large	$0.0094	24.0s	77%	90	86	83	81	81	84%
2	Mistral Large 2	$0.0090	23.4s	83%	85	84	84	84	83	84%
3	Mistral Medium 3.1	$0.0039	28.3s	81%	86	85	85	84	80	84%
17	Writer: Palmyra X5	$0.010	20.7s	73%	90	88	84	82	72	83%
12	ByteDance Seed 1.6 Flash	$0.0011	23.5s	74%	92	84	81	79	79	83%
26	Claude Sonnet 4.5	$0.031	38.6s	76%	89	82	81	81	81	83%
11	Z.AI GLM 5	$0.0068	1.0m	77%	87	85	82	82	78	83%
21	o4 Mini High	$0.023	50.3s	77%	87	84	81	80	80	82%
45	Rocinante 12B	$0.0008	19.4s	67%	92	91	80	78	71	82%
77	Claude Opus 4	$0.137	59.1s	77%	85	85	81	81	80	82%
27	Gemini 2.5 Pro	$0.031	32.1s	77%	86	84	81	81	78	82%
73	Gemini 3.1 Pro (Preview)	$0.098	1.6m	79%	84	83	81	81	81	82%
19	MoonshotAI: Kimi K2.5	$0.014	1.1m	77%	86	85	83	80	76	82%
14	Hermes 3 405B	$0.0019	37.5s	75%	88	85	84	79	74	82%
43	GPT-5.2	$0.054	1.5m	80%	83	82	81	81	81	82%
10	Z.AI GLM 4.6	$0.0056	1.0m	78%	85	83	81	81	79	82%
13	DeepSeek V3 (2025-03-24)	$0.0009	50.3s	76%	87	83	83	80	75	82%
6	Mistral Small Creative	$0.0005	6.9s	77%	85	83	83	82	76	82%
9	Mistral Large 3	$0.0024	24.5s	78%	84	83	82	79	78	81%
33	WizardLM 2 8x22b	$0.0019	2.2m	76%	86	82	81	81	76	81%
15	Gemma 3 27B	$0.0004	52.4s	77%	84	82	80	80	77	81%
16	Grok 4,1 Fast	$0.0016	34.8s	76%	84	83	80	79	78	81%
20	GPT-5 Mini	$0.010	1.3m	79%	82	82	81	79	78	81%
34	GPT-4.1	$0.018	50.5s	76%	85	82	82	80	73	80%
35	Claude Haiku 4.5	$0.0091	19.3s	72%	87	85	83	78	69	80%
70	Claude Opus 4.5	$0.071	59.5s	76%	84	81	80	79	76	80%
52	Claude Sonnet 4	$0.026	40.3s	72%	87	80	78	77	77	80%
25	DeepSeek V3.1	$0.0014	1.5m	77%	83	82	81	77	76	80%
49	DeepSeek V3.2	$0.0011	2.2m	72%	87	82	78	76	76	80%
61	ByteDance Seed 1.6	$0.015	3.0m	73%	85	84	81	79	71	80%
18	Claude 3 Haiku	$0.0022	19.2s	76%	83	81	80	77	75	79%
22	Ministral 3 8B	$0.0003	5.1s	74%	83	81	79	78	76	79%
76	Gemini 3 Pro (Preview)	$0.057	56.9s	72%	87	79	79	77	73	79%
72	Grok 4	$0.047	1.8m	74%	84	82	81	76	72	79%
39	DeepSeek V3 (2024-12-26)	$0.0017	1.1m	74%	83	81	80	79	72	79%
60	Claude 3.7 Sonnet	$0.032	39.3s	72%	87	80	79	76	73	79%
23	Gemini 2.5 Flash	$0.0051	11.5s	76%	82	81	81	77	73	79%
32	Minimax M2.5	$0.0028	1.2m	77%	80	80	80	77	76	79%
40	GPT-4o, Aug. 6th (temp=0)	$0.016	17.3s	75%	81	81	79	76	74	78%
29	Hermes 3 70B	$0.0006	34.2s	75%	81	81	80	75	75	78%
36	Ministral 8B	$0.0002	6.3s	72%	85	79	78	78	72	78%
42	Qwen 3.5 Plus (2026-02-15)	$0.0053	33.7s	73%	84	78	78	77	74	78%
31	Arcee AI: Trinity Large (Preview)	$0.0000	31.0s	75%	81	79	79	77	75	78%
47	Z.AI GLM 4.7	$0.0097	1.3m	75%	81	79	78	78	75	78%
28	Ministral 3 14B	$0.0004	8.0s	75%	81	79	78	77	74	78%
51	Z.AI GLM 4.5	$0.0038	43.1s	71%	84	82	80	72	70	78%
74	Claude 3.5 Sonnet	$0.037	32.4s	70%	85	80	77	76	71	78%
58	GPT-4o, May 13th (temp=0)	$0.029	11.4s	72%	84	77	77	75	74	77%
44	Gemini 3 Flash (Preview)	$0.0087	23.0s	74%	80	80	78	75	73	77%
67	GPT-4o, Aug. 6th (temp=1)	$0.016	19.8s	69%	86	80	78	71	70	77%
38	GPT-4.1 Mini	$0.0027	22.0s	74%	79	78	77	76	75	77%
56	GPT-4o, May 13th (temp=1)	$0.029	14.5s	74%	79	79	77	76	73	77%
46	GPT-4.1 Nano	$0.0007	15.4s	72%	81	77	76	75	74	77%
50	Gemini 2.5 Flash Lite	$0.0008	8.1s	70%	83	77	77	74	71	76%
57	Grok 4 Fast	$0.0015	25.2s	70%	81	80	75	73	72	76%
66	Llama 3.1 70B	$0.0015	27.1s	68%	83	80	79	75	62	76%
41	Ministral 3B	$0.0001	3.0s	74%	78	77	77	74	73	76%
54	Arcee AI: Trinity Mini	$0.0002	8.5s	70%	82	78	76	74	69	76%
48	Mistral NeMO	$0.0004	7.5s	71%	79	78	75	73	72	76%
55	Z.AI GLM 4.7 Flash	$0.0017	1.2m	73%	78	77	76	74	72	76%
64	Gemma 3 12B	$0.0002	39.9s	69%	82	78	74	72	72	76%
53	GPT-4o Mini (temp=1)	$0.0010	32.6s	72%	79	79	77	73	70	76%
62	DeepSeek-V2 Chat	$0.0017	1.4m	71%	80	76	76	73	71	75%
65	Gemma 3 4B	$0.0002	21.2s	69%	77	77	75	74	67	74%
63	Llama 3.1 8B	$0.0002	1.6m	73%	75	75	75	74	73	74%
69	Qwen 2.5 72B	$0.0007	26.7s	69%	79	75	74	72	69	74%
59	Ministral 3 3B	$0.0002	2.7s	71%	77	75	74	73	70	74%
80	Mistral Small 3.2 24B	$0.010	9.7m	65%	83	72	72	72	68	73%
68	Llama 3.1 Nemotron 70B	$0.0025	29.5s	71%	75	74	73	72	71	73%
71	GPT-4o Mini (temp=0)	$0.0010	36.5s	70%	77	73	73	71	71	73%
75	Stealth: Aurora Alpha	$0.0000	9.5s	69%	74	73	73	72	66	72%
78	GPT-5 Nano	$0.0039	1.2m	65%	74	70	69	67	64	69%
79	Claude 3.5 Haiku	$0.0027	11.3s	62%	73	72	69	69	58	68%
79.21%

Median	Evaluator	Top 3	Flop 3
99.5%	"Not X but Y" pattern overuse	100DeepSeek-V2 Chat 100Mistral Large 2 100Z.AI GLM 4.7	1GPT-5 Nano 31Grok 4 Fast 59Mistral Small Creative
33.4%	Adverb-first sentence starts	98Mistral Medium 3.1 94Mistral Small Creative 80Writer: Palmyra X5	0Stealth: Aurora Alpha 0GPT-4o, May 13th (temp=0) 0Z.AI GLM 4.7
100.0%	Adverbs in dialogue tags	100Ministral 8B 100Z.AI GLM 4.5 100Qwen 3.5 Plus (2026-02-15)	40Hermes 3 70B 59Claude Haiku 4.5 60Ministral 3B
88.5%	AI-ism adverb frequency	100ByteDance Seed 1.6 Flash 100Grok 4,1 Fast 100MoonshotAI: Kimi K2.5	71GPT-4.1 Nano 76Claude Haiku 4.5 76Gemma 3 12B
96.0%	AI-ism character names	100Grok 4 100GPT-5 Mini 100Stealth: Aurora Alpha	68Claude Opus 4 72Claude Opus 4.5 76Claude Sonnet 4.5
100.0%	AI-ism location names	100Gemini 3.1 Pro (Preview) 100Rocinante 12B 100Hermes 3 405B	—
38.9%	AI-ism word frequency	87Claude Sonnet 4.6 79GPT-5 77GPT-5 Mini	0GPT-4o, Aug. 6th (temp=1) 1GPT-4o Mini (temp=0) 4Claude 3.5 Haiku
100.0%	Cliché density	100Ministral 8B 100GPT-5 Nano 100Mistral Large 2	40GPT-4o, May 13th (temp=0) 60DeepSeek-V2 Chat 60Llama 3.1 8B
78.9%	Dialogue tag variety (said vs. fancy)	100DeepSeek V3.2 100Claude Opus 4.5 100GPT-5.1	0Claude 3.5 Haiku 1GPT-4o, Aug. 6th (temp=1) 9GPT-4o, May 13th (temp=1)
42.5%	Em-dash & semicolon overuse	100Mistral Small 3.2 24B 100Qwen 3.5 397B A17B 100Gemini 3.1 Pro (Preview)	0Claude Opus 4.6 0GPT-5 Nano 0GPT-4.1
100.0%	Emotion telling (show vs. tell)	100GPT-4.1 Mini 100GPT-4.1 100Gemini 2.5 Flash	80Mistral Small 3.2 24B 87Hermes 3 70B 94Claude 3 Haiku
95.7%	Filter word density	100Claude Sonnet 4.6 100o4 Mini High 100Ministral 8B	30Stealth: Aurora Alpha 42Claude 3.5 Sonnet 46Llama 3.1 Nemotron 70B
100.0%	Gibberish response detection	100Z.AI GLM 4.7 Flash 100Gemma 3 27B 100Llama 3.1 Nemotron 70B	60Llama 3.1 70B 65Llama 3.1 8B 80DeepSeek V3 (2025-03-24)
100.0%	Markdown formatting overuse	100Claude Opus 4.5 100Gemma 3 12B 100Grok 4	80Rocinante 12B 80Ministral 3 3B
100.0%	Missing dialogue indicators (quotation marks)	100GPT-4o Mini (temp=0) 100GPT-5 Mini 100Stealth: Aurora Alpha	80o4 Mini 95Hermes 3 405B 97Gemma 3 12B
34.6%	Name drop frequency	97Claude Sonnet 4.6 91Gemini 2.5 Pro 90ByteDance Seed 1.6	0GPT-4o Mini (temp=0) 2Z.AI GLM 4.5 3Claude 3 Haiku
83.0%	Narrator intent-glossing	100GPT-4.1 100GPT-4o Mini (temp=1) 100GPT-5.1	10GPT-5 Nano 36Claude Opus 4 48Stealth: Aurora Alpha
100.0%	Overuse of "that" (subordinate clause padding)	100Mistral Medium 3.1 100DeepSeek-V2 Chat 100Mistral Large 2	91Claude Haiku 4.5 95Claude 3.5 Haiku 95Z.AI GLM 4.5
100.0%	Paragraph length variance	100Z.AI GLM 4.7 100Writer: Palmyra X5 100Qwen 3.5 Plus (2026-02-15)	62Mistral Small 3.2 24B 70GPT-5 Nano 82Stealth: Aurora Alpha
94.3%	Passive voice overuse	100GPT-5.1 100Grok 4 Fast 100o4 Mini	84GPT-4o, Aug. 6th (temp=0) 85Z.AI GLM 4.6 85Z.AI GLM 4.7
100.0%	Past progressive (was/were + -ing) overuse	100GPT-5 Mini 100Qwen 3.5 Plus (2026-02-15) 100GPT-4o Mini (temp=1)	6Z.AI GLM 4.7 Flash 56Z.AI GLM 4.6 57DeepSeek V3.2
96.1%	Pronoun-first sentence starts	100GPT-5.2 100Ministral 3 14B 100DeepSeek V3 (2025-03-24)	28Mistral Small 3.2 24B 37Mistral NeMO 49ByteDance Seed 1.6
98.0%	Purple prose (modifier overload)	100Stealth: Aurora Alpha 100Gemini 2.5 Pro 100GPT-4o Mini (temp=0)	87Mistral Small 3.2 24B 90Gemma 3 12B 91Gemma 3 4B
100.0%	Repeated phrase echo	100Claude Haiku 4.5 100Gemini 2.5 Flash Lite 100Ministral 3 3B	—
100.0%	Sentence length variance	100Mistral NeMO 100Gemini 2.5 Pro 100GPT-4o, May 13th (temp=0)	80Mistral Small 3.2 24B 95Stealth: Aurora Alpha 97Llama 3.1 8B
64.7%	Sentence opener variety	93Claude 3.5 Sonnet 89Grok 4,1 Fast 88Claude Opus 4	34GPT-5 Nano 39Stealth: Aurora Alpha 39Mistral Small 3.2 24B
27.4%	Subject-first sentence starts	88Writer: Palmyra X5 83Rocinante 12B 75Cohere Command R+ (Aug. 2024)	0Stealth: Aurora Alpha 0Qwen 2.5 72B 1Z.AI GLM 4.5
20.0%	Subordinate conjunction sentence starts	77ByteDance Seed 1.6 66Writer: Palmyra X5 60Mistral NeMO	0DeepSeek V3 (2024-12-26) 0Grok 4,1 Fast 0Gemini 3.1 Pro (Preview)
83.4%	Technical jargon density	100DeepSeek V3 (2024-12-26) 100Mistral Large 100o4 Mini High	0GPT-5 Nano 25Claude Opus 4.5 32Z.AI GLM 5
51.4%	Useless dialogue additions	100GPT-5 100ByteDance Seed 1.6 100Qwen 3.5 397B A17B	0Qwen 2.5 72B 0Gemma 3 4B 0GPT-4o Mini (temp=0)

Bad Writing Habits

Mystery: examining a crime scene

Performance Score Distribution (Top 20)

Price-Performance Score Distribution (Top 20)

Most Stable Models (Top 20)

Top Overall Models (Top 20)