Short text (~524 words), small codex (11 detailed entries)

Codex Red Herring (False Positive Detection)

Tests whether models correctly report "no violations" when a codex is fully consistent with the prose passage. Models that hallucinate false violations (false positives) fail. Uses a 2×2 matrix of text length × codex size, with bare and detailed-entry variants.

Hallucination

Performance Score Distribution (Top 20)

Click a model name to view its detail page.

	Score
GPT-5 Mini	100%
Z.AI GLM 5	100%
Claude Sonnet 4.6	100%
ByteDance Seed 1.6	100%
o4 Mini High	100%
Claude Opus 4.5	100%
Grok 4.1 Fast	100%
GPT-4.1	100%
Grok 4	100%
Gemini 2.5 Flash (Reasoning)	100%
Gemini 2.5 Flash Lite (Reasoning)	100%
Mistral Large 3	100%
Mistral Large 2	100%
Mistral Large	100%
Ministral 3 14B	100%
Ministral 3 8B	100%
Ministral 8B	100%
Claude Haiku 4.5	95%
GPT-4.1 Mini	93%
GPT-5.1	93%

	Score	Cost	Time
Ministral 8B	100%	$0.0004	338ms
Mistral Small 3.2 24B	63%	$0.0004	1.7s
Ministral 3 8B	100%	$0.0006	339ms
GPT-4.1 Mini	93%	$0.0007	1.6s
Ministral 3 14B	100%	$0.0008	531ms
Arcee AI: Trinity Mini	93%	$0.0013	41.5s
ByteDance Seed 1.6 Flash	84%	$0.0006	7.9s
Mistral Large 3	100%	$0.0020	774ms
Grok 4.1 Fast	100%	$0.0011	8.6s
GPT-4.1	100%	$0.0035	613ms
Grok 4 Fast	93%	$0.0013	13.5s
Hermes 3 405B	74%	$0.0039	2.5s
Gemini 2.5 Flash Lite (Reasoning)	100%	$0.0021	15.1s
Claude Haiku 4.5	95%	$0.0044	1.3s
Gemini 2.5 Flash (Reasoning)	100%	$0.0060	21.5s
Mistral Large 2	100%	$0.0078	650ms
Mistral Large	100%	$0.0078	1.0s
Minimax M2.5	78%	$0.0029	22.7s
ByteDance Seed 1.6	100%	$0.0035	29.7s
Cohere Command R+ (Aug. 2024)	82%	$0.010	2.1s

Rank	Model	Avg. Cost	Avg. Time	Stability	# 1	# 2	# 3	# 4	# 5	# 6	# 7	# 8	# 9	# 10	Total
13	GPT-5 Mini	$0.0047	29.0s	100%	100	100	100	100	100	100	100	100	100	100	100%
15	Z.AI GLM 5	$0.012	57.7s	100%	100	100	100	100	100	100	100	100	100	100	100%
10	Claude Sonnet 4.6	$0.013	1.0s	100%	100	100	100	100	100	100	100	100	100	100	100%
12	ByteDance Seed 1.6	$0.0035	29.7s	100%	100	100	100	100	100	100	100	100	100	100	100%
16	o4 Mini High	$0.023	51.7s	100%	100	100	100	100	100	100	100	100	100	100	100%
14	Claude Opus 4.5	$0.021	2.0s	100%	100	100	100	100	100	100	100	100	100	100	100%
6	Grok 4.1 Fast	$0.0011	8.6s	100%	100	100	100	100	100	100	100	100	100	100	100%
5	GPT-4.1	$0.0035	613ms	100%	100	100	100	100	100	100	100	100	100	100	100%
19	Grok 4	$0.043	52.2s	100%	100	100	100	100	100	100	100	100	100	100	100%
11	Gemini 2.5 Flash (Reasoning)	$0.0060	21.5s	100%	100	100	100	100	100	100	100	100	100	100	100%
7	Gemini 2.5 Flash Lite (Reasoning)	$0.0021	15.1s	100%	100	100	100	100	100	100	100	100	100	100	100%
4	Mistral Large 3	$0.0020	774ms	100%	100	100	100	100	100	100	100	100	100	100	100%
8	Mistral Large 2	$0.0078	650ms	100%	100	100	100	100	100	100	100	100	100	100	100%
9	Mistral Large	$0.0078	1.0s	100%	100	100	100	100	100	100	100	100	100	100	100%
3	Ministral 3 14B	$0.0008	531ms	100%	100	100	100	100	100	100	100	100	100	100	100%
2	Ministral 3 8B	$0.0006	339ms	100%	100	100	100	100	100	100	100	100	100	100	100%
1	Ministral 8B	$0.0004	338ms	100%	100	100	100	100	100	100	100	100	100	100	100%
17	Claude Haiku 4.5	$0.0044	1.3s	70%	100	100	100	100	100	100	100	100	100	50	95%
18	GPT-4.1 Mini	$0.0007	1.6s	59%	100	100	100	100	100	100	100	100	100	32	93%
24	GPT-5.1	$0.022	27.2s	55%	100	100	100	100	100	100	100	100	100	25	93%
26	Aion 2.0	$0.0088	1.4m	55%	100	100	100	100	100	100	100	100	100	25	93%
23	o4 Mini	$0.012	22.4s	55%	100	100	100	100	100	100	100	100	100	25	93%
20	Grok 4 Fast	$0.0013	13.5s	55%	100	100	100	100	100	100	100	100	100	25	93%
22	GPT-5 Nano	$0.0023	43.0s	55%	100	100	100	100	100	100	100	100	100	25	93%
21	Arcee AI: Trinity Mini	$0.0013	41.5s	55%	100	100	100	100	100	100	100	100	100	25	93%
29	Claude Opus 4.6	$0.036	11.3s	49%	100	100	100	100	100	100	100	100	50	25	88%
44	Claude Opus 4.6 (Reasoning)	$0.079	37.1s	40%	100	100	100	100	100	100	100	100	25	25	85%
28	GPT-5.2	$0.013	16.3s	40%	100	100	100	100	100	100	100	100	25	25	85%
27	Z.AI GLM 4.6	$0.0038	29.3s	40%	100	100	100	100	100	100	100	100	25	25	85%
33	Z.AI GLM 4.7	$0.018	1.5m	40%	100	100	100	100	100	100	100	100	25	25	85%
25	ByteDance Seed 1.6 Flash	$0.0006	7.9s	35%	100	100	100	100	100	100	100	100	25	13	84%
30	Cohere Command R+ (Aug. 2024)	$0.010	2.1s	27%	100	100	100	100	100	100	100	100	10	8	82%
81	Gemini 3 Pro (Preview)	$0.118	1.3m	31%	100	100	100	100	100	100	100	25	25	25	78%
31	Minimax M2.5	$0.0029	22.7s	31%	100	100	100	100	100	100	100	25	25	25	78%
34	Z.AI GLM 4.7 Flash	$0.0024	1.5m	31%	100	100	100	100	100	100	100	25	25	25	78%
32	Hermes 3 405B	$0.0039	2.5s	21%	100	100	100	100	100	100	100	17	13	13	74%
36	Mistral Small 3.2 24B	$0.0004	1.7s	8%	100	100	100	100	100	100	8	6	6	5	63%
78	MoonshotAI: Kimi K2.5	$0.017	3.9m	16%	100	100	100	100	100	25	25	25	25	25	63%
50	Gemini 3 Flash (Preview, Reasoning)	$0.024	43.4s	16%	100	100	100	100	100	25	25	25	25	25	63%
75	Gemini 2.5 Pro	$0.069	49.8s	16%	100	100	100	100	100	25	25	25	25	25	63%
35	Llama 3.1 70B	$0.0019	3.5s	14%	100	100	100	100	42	42	38	25	25	25	60%
37	GPT-4o, Aug. 6th (temp=1)	$0.0065	1.1s	13%	100	100	100	100	42	42	25	25	25	25	58%
40	Z.AI GLM 4.5	$0.0032	21.3s	9%	100	100	100	100	100	29	29	13	4	3	58%
79	Claude Sonnet 4.6 (Reasoning)	$0.065	51.6s	7%	100	100	100	100	25	25	25	25	25	25	55%
77	GPT-5	$0.045	1.3m	7%	100	100	100	100	25	25	25	25	25	25	55%
42	GPT-4o, Aug. 6th (temp=0)	$0.0055	1.2s	6%	100	100	100	100	25	25	25	25	25	17	54%
84	Qwen 3.5 397B A17B	$0.024	5.4m	6%	100	100	100	100	25	25	25	25	17	17	53%
39	Mistral Medium 3.1	$0.0018	2.6s	13%	100	100	100	50	42	32	25	25	25	17	52%
57	Claude Sonnet 4	$0.014	2.9s	3%	100	100	100	100	17	17	17	17	17	17	50%
41	GPT-4.1 Nano	$0.0004	4.9s	10%	100	100	100	30	29	28	28	27	27	27	50%
53	Rocinante 12B	$0.0009	8.3s	2%	100	100	100	100	25	10	6	4	4	3	45%
45	Hermes 3 70B	$0.0012	8.3s	7%	100	100	100	50	31	25	17	13	10	6	45%
66	Claude 3.5 Sonnet	$0.031	6.5s	4%	100	100	100	25	17	17	17	17	13	10	41%
83	Gemini 3.1 Pro (Preview)	$0.084	58.8s	10%	100	100	25	25	25	25	25	25	25	25	40%
48	Mistral Small Creative	$0.0005	2.6s	10%	100	100	35	29	29	29	29	17	8	7	38%
60	Claude 3.7 Sonnet	$0.015	3.7s	7%	100	100	38	35	25	17	13	13	13	13	36%
82	Claude Opus 4	$0.071	5.2s	7%	100	100	25	25	25	17	17	17	17	17	36%
46	Qwen 2.5 72B	$0.0007	10.8s	17%	100	31	31	30	30	29	29	25	25	25	35%
38	Claude 3 Haiku	$0.0014	2.9s	31%	42	42	42	38	38	38	35	35	35	10	35%
52	DeepSeek V3 (2025-03-24)	$0.0008	4.3s	14%	100	25	25	25	25	25	25	25	25	25	33%
43	Ministral 3B	$0.0002	3.8s	23%	50	42	33	32	30	29	29	29	29	4	31%
70	Arcee AI: Trinity Large (Preview)	$0.0000	2.0m	21%	50	35	28	27	27	27	27	26	26	4	28%
49	Llama 3.1 Nemotron 70B	$0.0051	12.9s	26%	38	38	38	35	35	35	33	17	6	4	28%
80	Qwen 3.5 Plus (2026-02-15)	$0.023	2.2m	21%	50	42	38	35	33	25	17	13	10	10	27%
54	Ministral 3 3B	$0.0006	8.5s	21%	42	38	33	29	28	28	27	26	3	1	26%
47	DeepSeek V3 (2024-12-26)	$0.0014	2.4s	25%	25	25	25	25	25	25	25	25	25	25	25%
55	GPT-4o Mini (temp=1)	$0.0007	6.7s	23%	35	33	32	31	31	30	29	5	4	4	23%
51	DeepSeek-V2 Chat	$0.0013	2.1s	23%	25	25	25	25	25	25	25	25	17	17	23%
56	DeepSeek V3.1	$0.0008	8.0s	23%	25	25	25	25	25	25	25	25	13	13	23%
58	Llama 3.1 8B	$0.0002	31.4s	21%	38	30	29	28	28	27	27	8	6	4	22%
67	GPT-4o, May 13th (temp=1)	$0.023	3.3s	16%	42	35	33	29	25	17	17	13	10	4	22%
62	DeepSeek V3.2	$0.0010	14.7s	14%	38	33	25	17	17	17	17	17	13	10	20%
64	WizardLM 2 8x22b	$0.0027	26.8s	15%	42	25	25	25	25	13	13	10	8	4	19%
63	Writer: Palmyra X5	$0.0038	5.3s	15%	25	25	25	25	17	17	17	10	10	10	18%
68	Claude Sonnet 4.5	$0.016	4.9s	15%	25	25	17	17	17	17	17	17	13	10	17%
59	Gemini 2.5 Flash	$0.0013	1.7s	17%	25	25	25	25	25	17	13	13	3	2	17%
61	Gemini 3 Flash (Preview)	$0.0017	2.3s	15%	25	17	17	17	17	17	17	17	10	10	16%
72	GPT-4o, May 13th (temp=0)	$0.023	2.4s	14%	32	17	17	17	17	17	10	8	7	7	15%
69	Gemini 2.5 Flash Lite	$0.0006	2.2s	2%	100	6	6	5	4	4	3	3	3	3	13%
65	Gemma 3 12B	$0.0003	5.8s	12%	17	17	13	13	13	13	13	8	7	6	12%
71	Gemma 3 27B	$0.0005	9.5s	7%	13	10	8	8	7	7	6	6	6	5	8%
73	Mistral NeMO	$0.0007	4.7s	5%	8	6	6	5	5	5	4	4	4	3	5%
74	Gemma 3 4B	$0.0002	6.2s	4%	6	6	5	5	5	5	5	4	4	3	5%
76	GPT-4o Mini (temp=0)	$0.0007	9.9s	4%	4	4	4	4	4	4	4	4	4	4	4%
59.16%

Median	Evaluator	Top 3	Flop 3
50.0%	Correct "no violations" response	100Grok 4.1 Fast 100Gemini 2.5 Flash Lite (Reasoning) 100Mistral Large	0Gemma 3 12B 0Gemini 2.5 Flash 0Claude Sonnet 4.5
67.5%	No hallucinated violations	100Gemini 2.5 Flash Lite (Reasoning) 100Ministral 3 14B 100Grok 4	8GPT-4o Mini (temp=0) 9Gemma 3 4B 10Mistral NeMO

Codex Red Herring (False Positive Detection)

Short text (~524 words), small codex (11 detailed entries)

Performance Score Distribution (Top 20)

Price-Performance Score Distribution (Top 20)

Most Stable Models (Top 20)

Top Overall Models (Top 20)