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.

Model	Total ▼	Literary fiction: old friends reunite	Thriller: chase through city streets	Romance: separated couple reunites	Fantasy: entering an ancient ruin	Mystery: examining a crime scene	Horror: alone in an eerie place at night
Rocinante 12B	87%	88%	90%	86%	84%	88%	87%
Mistral Large	87%	86%	87%	85%	85%	89%	88%
Mistral Large 2	87%	86%	89%	87%	86%	86%	86%
Ministral 8B	86%	86%	89%	83%	84%	84%	90%
Ministral 3B	86%	82%	89%	84%	84%	87%	89%
Claude 3.5 Sonnet	84%	84%	85%	86%	80%	87%	84%
Mistral NeMO	84%	87%	80%	85%	80%	84%	85%
Claude 3.7 Sonnet	83%	80%	87%	79%	84%	85%	85%
DeepSeek-V2 Chat	83%	80%	87%	81%	82%	85%	84%
Hermes 3 70B	83%	81%	83%	83%	82%	85%	84%
Hermes 3 405B	83%	80%	83%	81%	82%	87%	82%
Llama 3.1 70B	83%	85%	83%	87%	75%	87%	79%
Qwen 2.5 72B	83%	84%	80%	81%	81%	83%	85%
Llama 3.1 8B	83%	81%	89%	79%	78%	83%	85%
Cohere Command R+ (Aug. 2024)	82%	81%	82%	84%	79%	85%	82%

1–15 of 25

Page 1 / 2

Rows:

Cost vs Performance

Compares total cost for this test against the test score. Quadrant lines are drawn at the median values. Only models with available cost data are shown.

Literary fiction: old friends reunite

0-shot Creative writingRule following

Performance Score Distribution (Top 20)

Click a model name to view its detail page.

Rank	Model	Avg. Cost	Avg. Time	Stability	# 1	# 2	# 3	# 4	# 5	Total
4	Rocinante 12B	$0.0010	31.8s	79%	92	92	86	85	83	88%
2	Mistral NeMO	$0.0004	19.7s	82%	91	89	87	86	83	87%
1	Stealth: Aurora Alpha	$0.0000	10.3s	82%	89	86	83	–	–	86%
3	Ministral 8B	$0.0003	12.7s	81%	89	88	86	86	81	86%
7	Mistral Large 2	$0.013	34.9s	78%	93	90	87	81	79	86%
6	Mistral Large	$0.012	33.8s	82%	89	88	86	86	82	86%
11	Llama 3.1 70B	$0.0023	54.6s	74%	96	86	83	82	80	85%
5	Qwen 2.5 72B	$0.0008	33.9s	82%	86	85	85	82	81	84%
25	Claude 3.5 Sonnet	$0.048	41.3s	74%	92	85	81	80	80	84%
9	Ministral 3B	$0.0001	7.0s	73%	90	82	80	79	77	82%
17	Cohere Command R+ (Aug. 2024)	$0.020	52.5s	79%	84	83	83	79	79	81%
15	GPT-4o, Aug. 6th (temp=0)	$0.018	24.5s	76%	86	82	81	81	76	81%
10	Llama 3.1 8B	$0.0002	41.3s	77%	84	83	80	80	78	81%
8	Llama 3.1 Nemotron 70B	$0.0026	41.8s	79%	82	82	81	81	79	81%
16	GPT-4o, May 13th (temp=0)	$0.026	13.1s	77%	85	82	80	80	78	81%
18	Hermes 3 70B	$0.0007	55.2s	75%	85	84	81	78	76	81%
14	WizardLM 2 8x22b	$0.0023	42.7s	76%	84	82	81	80	76	81%
13	DeepSeek-V2 Chat	$0.0019	47.3s	77%	83	83	81	78	78	80%
24	Hermes 3 405B	$0.0020	1.6m	74%	86	83	80	76	75	80%
19	GPT-4o Mini (temp=0)	$0.0012	27.7s	71%	86	85	79	78	72	80%
12	GPT-4o Mini (temp=1)	$0.0012	31.4s	76%	84	82	81	78	75	80%
23	Claude 3.7 Sonnet	$0.040	48.3s	78%	82	81	81	80	76	80%
22	GPT-4o, May 13th (temp=1)	$0.026	13.0s	73%	86	79	79	77	76	79%
20	GPT-4o, Aug. 6th (temp=1)	$0.018	34.1s	77%	80	79	78	78	78	79%
21	Claude 3 Haiku	$0.0022	15.6s	71%	83	79	77	76	71	77%
82.27%

Thriller: chase through city streets

0-shot Creative writingRule following

Performance Score Distribution (Top 20)

Click a model name to view its detail page.

Rank	Model	Avg. Cost	Avg. Time	Stability	# 1	# 2	# 3	# 4	# 5	Total
13	Rocinante 12B	$0.0031	2.6m	80%	98	92	88	86	84	90%
3	Mistral Large 2	$0.011	28.8s	85%	92	90	89	89	86	89%
2	Ministral 3B	$0.0001	5.6s	84%	93	91	89	87	85	89%
1	Ministral 8B	$0.0002	8.4s	85%	93	89	89	87	86	89%
7	Llama 3.1 8B	$0.0003	59.7s	75%	98	96	90	81	79	89%
4	Mistral Large	$0.012	33.0s	84%	91	89	89	85	83	87%
6	DeepSeek-V2 Chat	$0.0015	33.0s	76%	96	91	87	86	76	87%
12	WizardLM 2 8x22b	$0.0018	24.7s	72%	100	95	94	78	68	87%
14	Claude 3.7 Sonnet	$0.037	46.5s	82%	92	90	88	83	83	87%
19	Claude 3.5 Sonnet	$0.042	38.5s	79%	89	88	84	83	81	85%
11	Llama 3.1 Nemotron 70B	$0.0021	29.8s	77%	88	87	84	80	79	84%
5	GPT-4o Mini (temp=0)	$0.0011	26.3s	81%	86	84	83	82	82	83%
18	Hermes 3 405B	$0.0022	1.5m	74%	89	88	81	80	78	83%
9	Llama 3.1 70B	$0.0007	23.3s	77%	87	85	84	83	76	83%
16	Hermes 3 70B	$0.0006	45.2s	75%	90	85	84	82	74	83%
8	GPT-4o Mini (temp=1)	$0.0011	23.6s	79%	86	84	83	82	78	83%
25	Cohere Command R+ (Aug. 2024)	$0.021	2.1m	75%	89	85	82	80	76	82%
21	GPT-4o, Aug. 6th (temp=1)	$0.015	30.5s	73%	89	85	80	79	78	82%
22	GPT-4o, May 13th (temp=1)	$0.028	14.2s	76%	87	83	81	80	78	82%
24	GPT-4o, May 13th (temp=0)	$0.026	13.3s	73%	90	83	81	81	74	82%
15	Claude 3 Haiku	$0.0021	15.8s	75%	88	82	80	79	78	82%
17	Mistral NeMO	$0.0005	20.5s	73%	87	84	81	76	75	80%
10	Stealth: Aurora Alpha	$0.0000	16.4s	80%	81	81	80	–	–	80%
20	Qwen 2.5 72B	$0.0008	39.1s	72%	88	80	79	79	76	80%
23	GPT-4o, Aug. 6th (temp=0)	$0.016	21.9s	75%	84	80	80	80	74	80%
84.35%

Romance: separated couple reunites

0-shot Creative writingRule following

Performance Score Distribution (Top 20)

Click a model name to view its detail page.

Rank	Model	Avg. Cost	Avg. Time	Stability	# 1	# 2	# 3	# 4	# 5	Total
7	Mistral Large 2	$0.012	33.6s	77%	94	90	86	85	79	87%
3	Llama 3.1 70B	$0.0008	39.0s	78%	93	89	86	83	80	87%
14	Claude 3.5 Sonnet	$0.046	36.1s	81%	89	89	86	84	83	86%
8	Rocinante 12B	$0.0007	17.7s	74%	94	91	83	83	78	86%
2	Mistral Large	$0.011	28.6s	82%	87	87	85	85	82	85%
1	Mistral NeMO	$0.0003	9.8s	80%	88	88	84	83	81	85%
13	Cohere Command R+ (Aug. 2024)	$0.020	43.0s	78%	89	84	83	82	81	84%
4	Ministral 3B	$0.0001	4.9s	79%	88	84	83	81	81	84%
5	Ministral 8B	$0.0004	17.2s	79%	86	85	83	83	79	83%
6	Hermes 3 70B	$0.0006	49.8s	79%	86	83	83	82	80	83%
10	Qwen 2.5 72B	$0.0009	37.0s	76%	87	83	82	78	77	81%
15	GPT-4o, Aug. 6th (temp=1)	$0.017	27.8s	78%	84	83	82	81	76	81%
17	Hermes 3 405B	$0.0022	1.6m	76%	84	83	80	80	79	81%
9	Stealth: Aurora Alpha	$0.0000	8.6s	75%	86	80	78	–	–	81%
12	GPT-4o, Aug. 6th (temp=0)	$0.015	23.6s	78%	84	82	82	79	77	81%
20	GPT-4o, May 13th (temp=0)	$0.025	14.4s	75%	84	83	79	79	78	81%
16	DeepSeek-V2 Chat	$0.0018	50.7s	75%	87	82	81	78	76	81%
11	GPT-4o Mini (temp=0)	$0.0012	32.7s	77%	84	81	80	78	78	80%
25	Claude 3.7 Sonnet	$0.039	45.9s	73%	84	82	79	75	75	79%
24	Llama 3.1 8B	$0.0005	3.7m	76%	81	80	79	78	77	79%
22	GPT-4o, May 13th (temp=1)	$0.026	13.7s	72%	83	82	77	76	74	79%
18	GPT-4o Mini (temp=1)	$0.0012	29.9s	75%	80	79	79	76	75	78%
21	Llama 3.1 Nemotron 70B	$0.0023	35.4s	73%	82	78	76	75	74	77%
19	WizardLM 2 8x22b	$0.0016	24.8s	74%	79	78	77	75	74	77%
23	Claude 3 Haiku	$0.0020	15.2s	70%	78	77	74	72	70	74%
81.55%

Fantasy: entering an ancient ruin

0-shot Creative writingRule following

Performance Score Distribution (Top 20)

Click a model name to view its detail page.

Rank	Model	Avg. Cost	Avg. Time	Stability	# 1	# 2	# 3	# 4	# 5	Total
4	Mistral Large 2	$0.013	31.1s	77%	91	90	84	83	80	86%
5	Mistral Large	$0.014	34.6s	79%	89	88	86	82	78	85%
2	Ministral 3B	$0.0001	3.5s	78%	90	85	84	82	80	84%
16	Claude 3.7 Sonnet	$0.044	53.7s	80%	87	86	85	84	78	84%
1	Ministral 8B	$0.0004	11.7s	82%	86	85	84	84	82	84%
3	Rocinante 12B	$0.0009	26.7s	80%	87	86	84	81	79	84%
14	Hermes 3 70B	$0.0011	1.5m	76%	87	84	82	80	77	82%
17	Hermes 3 405B	$0.0030	1.6m	76%	89	83	83	78	78	82%
8	DeepSeek-V2 Chat	$0.0023	46.2s	79%	85	84	84	80	77	82%
11	GPT-4o, May 13th (temp=1)	$0.032	16.1s	78%	86	83	81	81	79	82%
9	GPT-4o, Aug. 6th (temp=0)	$0.017	17.4s	77%	86	84	83	81	74	82%
12	Qwen 2.5 72B	$0.0010	39.4s	72%	87	87	78	78	78	81%
7	GPT-4o Mini (temp=0)	$0.0011	28.9s	78%	83	82	81	79	78	81%
6	Mistral NeMO	$0.0005	7.9s	75%	86	82	82	76	75	80%
24	Claude 3.5 Sonnet	$0.047	38.6s	78%	82	81	80	78	78	80%
19	GPT-4o, May 13th (temp=0)	$0.034	16.9s	74%	84	83	79	76	75	80%
10	GPT-4o Mini (temp=1)	$0.0012	28.4s	74%	83	83	79	76	75	79%
22	Cohere Command R+ (Aug. 2024)	$0.021	51.5s	73%	84	81	78	77	75	79%
21	Llama 3.1 8B	$0.0002	1.2m	72%	82	82	77	76	75	78%
18	GPT-4o, Aug. 6th (temp=1)	$0.017	16.5s	73%	83	78	78	75	74	78%
15	Claude 3 Haiku	$0.0025	16.9s	71%	82	81	77	75	72	78%
13	Stealth: Aurora Alpha	$0.0000	10.8s	70%	81	75	74	–	–	77%
20	Llama 3.1 Nemotron 70B	$0.0031	32.6s	70%	81	81	76	74	72	77%
23	WizardLM 2 8x22b	$0.0021	23.8s	68%	80	79	73	72	71	75%
25	Llama 3.1 70B	$0.0010	37.9s	67%	82	79	73	72	69	75%
80.53%

Mystery: examining a crime scene

0-shot Creative writingRule following

Performance Score Distribution (Top 20)

Click a model name to view its detail page.

Rank	Model	Avg. Cost	Avg. Time	Stability	# 1	# 2	# 3	# 4	# 5	Total
1	Mistral Large	$0.013	35.0s	85%	92	91	89	88	87	89%
2	Rocinante 12B	$0.0006	25.4s	80%	95	90	88	85	83	88%
13	Hermes 3 405B	$0.0020	59.2s	78%	94	93	89	81	80	87%
25	Claude 3.5 Sonnet	$0.044	37.5s	72%	96	94	83	82	82	87%
3	Llama 3.1 70B	$0.0007	33.1s	83%	90	88	87	86	83	87%
10	Ministral 3B	$0.0001	1.1m	81%	91	88	86	85	84	87%
11	Mistral Large 2	$0.011	28.7s	80%	93	86	86	84	83	86%
9	Llama 3.1 Nemotron 70B	$0.0023	32.3s	81%	88	87	86	84	79	85%
14	Hermes 3 70B	$0.0006	31.1s	78%	90	89	84	81	80	85%
18	Cohere Command R+ (Aug. 2024)	$0.016	1.3m	82%	87	87	87	83	80	85%
6	DeepSeek-V2 Chat	$0.0016	36.9s	83%	86	86	85	84	83	85%
22	Claude 3.7 Sonnet	$0.037	46.9s	81%	88	86	85	83	81	85%
8	GPT-4o Mini (temp=0)	$0.0010	24.7s	81%	86	85	84	84	82	84%
4	Ministral 8B	$0.0002	9.9s	82%	86	86	85	82	82	84%
17	GPT-4o, Aug. 6th (temp=0)	$0.016	29.9s	77%	89	86	83	82	80	84%
20	GPT-4o, May 13th (temp=0)	$0.028	19.0s	77%	88	86	84	84	77	84%
5	Mistral NeMO	$0.0003	11.1s	81%	86	84	83	83	82	84%
7	GPT-4o Mini (temp=1)	$0.0011	25.2s	83%	85	84	84	83	82	84%
15	Qwen 2.5 72B	$0.0006	27.7s	75%	89	85	81	80	80	83%
19	GPT-4o, Aug. 6th (temp=1)	$0.016	36.9s	78%	87	86	84	79	78	83%
16	WizardLM 2 8x22b	$0.0016	24.6s	75%	88	87	83	80	75	83%
24	Llama 3.1 8B	$0.0003	2.0m	78%	85	85	83	81	78	83%
12	Stealth: Aurora Alpha	$0.0000	10.4s	78%	85	82	80	–	–	82%
23	GPT-4o, May 13th (temp=1)	$0.030	15.6s	78%	83	82	81	80	78	81%
21	Claude 3 Haiku	$0.0020	15.6s	75%	82	80	78	78	78	79%
84.51%

Horror: alone in an eerie place at night

0-shot Creative writingRule following

Performance Score Distribution (Top 20)

Click a model name to view its detail page.