Safety

Composite performance across 1 benchmarks · Normalized 0–100 · At least 1 scores required · Ranked by average performance · Top 50 of 2839 eligible models

#	Model	Provider	IFEval	Avg	Coverage	Price	Value
1	Meta: Llama 3.3 70B Instruct	Meta	90.0%	100.0	1/1	$1.38/M	72.5
2	MaziyarPanahi/calme-2.1-qwen2.5-72b	MaziyarPanahi	86.6%	96.3	1/1	—	—
3	Meta: Llama 3.1 70B Instruct	Meta	86.7%	96.3	1/1	$0.80/M	120.4
4	VAGOsolutions/Llama-3.1-SauerkrautLM-70b-Instruct	VAGOsolutions	86.6%	96.2	1/1	—	—
5	Qwen/Qwen2.5-72B-Instruct	Alibaba	86.4%	96.0	1/1	—	—
6	huihui-ai/Qwen2.5-72B-Instruct-abliterated	huihui-ai	85.9%	95.5	1/1	—	—
7	YOYO-AI/ZYH-LLM-Qwen2.5-14B-V3	YOYO-AI	85.8%	95.3	1/1	—	—
8	pankajmathur/orca_mini_v8_1_70b	pankajmathur	85.7%	95.3	1/1	—	—
9	dnhkng/RYS-Llama3.1-Large	dnhkng	84.9%	94.4	1/1	—	—
10	rubenroy/Gilgamesh-72B	rubenroy	84.9%	94.3	1/1	—	—
11	MaziyarPanahi/calme-2.2-qwen2.5-72b	MaziyarPanahi	84.8%	94.2	1/1	—	—
12	raphgg/test-2.5-72B	raphgg	84.4%	93.8	1/1	—	—
13	rombodawg/Rombos-LLM-V2.6-Qwen-14b	rombodawg	84.3%	93.7	1/1	—	—
14	ssmits/Qwen2.5-95B-Instruct	ssmits	84.3%	93.7	1/1	—	—
15	tensopolis/virtuoso-small-v2-tensopolis-v1	tensopolis	84.2%	93.6	1/1	—	—
16	Qwen/Qwen2.5-14B-Instruct-1M	Alibaba	84.1%	93.5	1/1	—	—
17	upstage/solar-pro-preview-instruct	Upstage	84.2%	93.5	1/1	—	—
18	mistralai/Mistral-Large-Instruct-2411	Mistral AI	84.0%	93.4	1/1	—	—
19	YOYO-AI/Qwen2.5-14B-1M-YOYO-V3	YOYO-AI	84.0%	93.3	1/1	—	—
20	YOYO-AI/Qwen2.5-14B-YOYO-V4	YOYO-AI	84.0%	93.3	1/1	—	—
21	LGAI-EXAONE/EXAONE-3.5-32B-Instruct	LG AI Research	83.9%	93.3	1/1	—	—
22	pankajmathur/orca_mini_v9_2_70b	pankajmathur	83.8%	93.2	1/1	—	—
23	allenai/Llama-3.1-Tulu-3-70B	Meta	83.8%	93.1	1/1	—	—
24	allenai/Llama-3.1-Tulu-3-70B	Ai2	83.8%	93.1	1/1	—	—
25	YOYO-AI/ZYH-LLM-Qwen2.5-14B-V4	YOYO-AI	83.6%	93.0	1/1	—	—
26	Qwen/Qwen2.5-32B-Instruct	Alibaba	83.5%	92.8	1/1	—	—
27	CombinHorizon/zetasepic-abliteratedV2-Qwen2.5-32B-Inst-BaseMerge-TIES	CombinHorizon	83.3%	92.6	1/1	—	—
28	huihui-ai/Qwen2.5-14B-Instruct-abliterated-v2	huihui-ai	83.3%	92.6	1/1	—	—
29	zetasepic/Qwen2.5-32B-Instruct-abliterated-v2	zetasepic	83.3%	92.6	1/1	—	—
30	vicgalle/Configurable-Llama-3.1-8B-Instruct	vicgalle	83.1%	92.4	1/1	—	—
31	Goekdeniz-Guelmez/Josiefied-Qwen2.5-14B-Instruct-abliterated-v4	Goekdeniz-Guelmez	82.9%	92.2	1/1	—	—
32	allenai/Llama-3.1-Tulu-3-70B-DPO	Ai2	82.8%	92.0	1/1	—	—
33	arcee-ai/Virtuoso-Small-v2	Arcee AI	82.7%	91.9	1/1	—	—
34	Saxo/Linkbricks-Horizon-AI-Avengers-V3-32B	Saxo	82.5%	91.7	1/1	—	—
35	allenai/Llama-3.1-Tulu-3-8B	Meta	82.5%	91.7	1/1	—	—
36	maldv/Awqward2.5-32B-Instruct	maldv	82.5%	91.7	1/1	—	—
37	allenai/Llama-3.1-Tulu-3-8B	Ai2	82.5%	91.7	1/1	—	—
38	CombinHorizon/Josiefied-abliteratedV4-Qwen2.5-14B-Inst-BaseMerge-TIES	CombinHorizon	82.4%	91.6	1/1	—	—
39	prithivMLmods/Dinobot-Opus-14B-Exp	prithivMLmods	82.4%	91.6	1/1	—	—
40	nbeerbower/llama3.1-kartoffeldes-70B	nbeerbower	82.3%	91.5	1/1	—	—
41	sthenno-com/miscii-14b-1028	sthenno	82.4%	91.5	1/1	—	—
42	tanliboy/lambda-qwen2.5-14b-dpo-test	tanliboy	82.3%	91.5	1/1	—	—
43	CombinHorizon/huihui-ai-abliterated-Qwen2.5-32B-Inst-BaseMerge-TIES	CombinHorizon	82.1%	91.2	1/1	—	—
44	YOYO-AI/Qwen2.5-14B-YOYO-V4-p1	YOYO-AI	82.0%	91.2	1/1	—	—
45	YOYO-AI/Qwen2.5-14B-it-restore	YOYO-AI	82.1%	91.2	1/1	—	—
46	MaziyarPanahi/calme-2.2-llama3-70b	MaziyarPanahi	82.1%	91.2	1/1	—	—
47	Saxo/Linkbricks-Horizon-AI-Avengers-V6-32B	Saxo	82.1%	91.2	1/1	—	—
48	v000000/Qwen2.5-14B-Gutenberg-Instruct-Slerpeno	v000000	82.0%	91.1	1/1	—	—
49	CombinHorizon/huihui-ai-abliteratedV2-Qwen2.5-14B-Inst-BaseMerge-TIES	CombinHorizon	81.8%	90.9	1/1	—	—
50	MaziyarPanahi/calme-2.1-qwen2-72b	MaziyarPanahi	81.6%	90.7	1/1	—	—

Meta: Llama 3.3 70B Instruct leads on raw performance, but Meta: Llama 3.1 70B Instruct delivers 66% more performance per dollar. That extra 3.7 points costs $0.16/M per point.

Access this data

JSON GET /api/v1

[
  {
    "rank": 1,
    "model": "Meta: Llama 3.3 70B Instruct",
    "provider": "Meta",
    "avg_score": 100.0,
    "cost": 1.38,
    "cost_unit": "million_tokens",
    "value": 72.5
  },
  {
    "rank": 2,
    "model": "MaziyarPanahi/calme-2.1-qwen2.5-72b",
    "provider": "MaziyarPanahi",
    "avg_score": 96.3,
    "cost": null,
    "cost_unit": "million_tokens",
    "value": null
  },
  {
    "rank": 3,
    "model": "Meta: Llama 3.1 70B Instruct",
    "provider": "Meta",
    "avg_score": 96.3,
    "cost": 0.8,
    "cost_unit": "million_tokens",
    "value": 120.4
  }
]

Curl

$ curl -s https://rightrank.ai/api/v1/rankings/safety.json | jq '.[:3]'

CLI brew install rightrank

$ rightrank top safety --limit 5
$ rightrank compare claude-opus-4.6 gpt-5.4-pro --category safety
$ rightrank value safety  # best performance per dollar

Claude Code Skill Skill

# In Claude Code, use the RightRank skill:
/rightrank best model for safety
/rightrank compare top 5 by value
/rightrank cheapest model scoring above 80 on safety