Which language models have the strongest safety evidence?

An evidence-weighted ranking of 818 language models across 112 independent evaluations. 268 models currently have enough breadth to rank.

Overall ranking

Rank combines model behavior (90%) and company governance (10%). Models need at least 3 evals across 2 behavior components. The interval is at least as important as the point estimate: it shows source-family sensitivity, not future performance.

How to interpret these ranks

The behavior order is the mean rank across complete-lineage bootstrap refits. Every sub-eval contributes only an ordering; native score scales are never pooled. Intervals separately resample transitive governed-source dependency components. Coverage varies and nearby models are often indistinguishable: the median 90% interval spans 140.5 ranks.

Rank by release date

258 of 268 ranked models have verified release dates. Each mark identifies the model's developer. Release date is context only and is not used to calculate rank.

What changed

Latest models with safety evidence

  1. Pokee-Isaac 28B v0 · 1 evals across 2 components
  2. Qwen3.8 Max · 2 evals across 3 components
  3. Inkling Small · 4 evals across 5 components
  4. Claude Opus 5 · 7 evals across 5 components
  5. Gemini 3.6 Flash · 5 evals across 5 components
  6. Gemini 3.5 Flash Lite · 7 evals across 5 components
  7. G9V3 3B · 1 evals across 1 components
  8. Kimi K3 · 13 evals across 6 components

Full ranking table

Rows are ordered by estimated rank, but broad overlapping intervals should not be read as evidence of a real difference.

268 models
Overall rank; brackets show the 90% source-sensitivity interval.
Estimated rank
[90% interval]
ModelEvalsComponentsRelease date
1 [1–6]claude-opus-575/72026-07-24
2 [2–15]claude-fable-5167/72026-06-09
3 [1–37]claude-264/72023-07-11
4 [3–47]claude-opus-4.8207/72026-05-28
5 [3–51]claude-opus-4.7257/72026-04-16
6 [3–62]claude-opus-4197/72025-05-22
7 [5–65]gpt-5.6-terra177/72026-07-09
8 [8–52]claude-opus-4.6237/72026-02-05
9 [9–73]gpt-5.5297/72026-04-23
10 [5–180]muse-spark-1.1117/72026-07-09
11 [6–56]gpt-5-nano18+27/72025-08-07
12 [3–87]gpt-5.3-chat55/72026-03-03
13 [4–72]claude-sonnet-4.6257/72026-01-21
14 [3–125]gpt-5.2-chat33/72025-12-11
15 [12–77]gpt-5.2267/72025-12-11
16 [7–96]gemini-3.1-pro-preview207/72026-02-19
17 [3–161]gpt-5.5-instant33/72026-05-26
18 [12–76]claude-haiku-4.5337/72025-10-15
19 [15–107]gpt-5-mini23+17/72025-08-07
20 [7–151]muse-spark33/72026-04-08
21 [5–106]claude-opus-4.5207/72025-11-24
22 [4–125]gemini-3.6-flash55/72026-07-21
23 [12–89]claude-sonnet-4.5327/72025-09-29
24 [10–95]claude-opus-4.1126/72025-08-05
25 [12–78]claude-3.7-sonnet257/72025-02-24
26 [9–91]gpt-5.4227/72026-03-05
27 [12–83]claude-sonnet-4317/72025-05-22
28 [16–220]qwen3.5-27b36/72026-02-24
29 [12–143]inkling-small2+25/72026-07-30
30 [10–137]grok-4.20-multi-agent44/72026-03-10
31 [7–166]phi-3.5-moe-instruct56/72024-08-22
32 [6–179]gpt-5.6-sol177/72026-07-09
33 [8–136]o1-preview46/72024-09-12
34 [4–193]gemini-2.5-pro-exp43/72025-03-25
35 [7–167]inkling5+26/72026-07-14
36 [13–140]gpt-4.5-preview96/72025-02-27
37 [9–208]llama-3.1-tulu-3-8b35/72024-11-20
38 [14–149]minimax-m35+15/72026-06-02
39 [5–177]mimo-v2.5-pro4+15/72026-04-27
40 [16–138]gpt-5327/72025-08-07
41 [2–234]gpt-5-pro34/72025-08-07
42 [33–131]qwen3.5-397b-a17b7+17/72026-02-16
43 [23–193]glm-4.5-air10+15/72025-07-28
44 [21–154]kimi-k3136/72026-07-16
45 [14–168]gpt-5.6-luna17+17/72026-07-09
46 [22–257]palmyra-x-00463/72024-10-09
47 [16–165]gemini-1.0-pro86/72023-12-06
48 [6–154]claude-3.5-haiku147/72024-10-22
49 [13–219]internlm2-chat-7b44/72024-01-17
50 [30–258]qwen3-235b-a22b-fp8-throughput63/72025-04-28
51 [27–143]qwen2-72b-instruct105/72024-06-07
52 [26–169]glm-59+17/72026-02-11
53 [37–181]deepseek-v4-flash13+17/72026-04-22
54 [18–216]internlm2-chat-20b45/72024-01-17
55 [26–189]qwen-72b-chat44/72023-11-30
56 [17–137]gpt-4.1257/72025-04-14
57 [36–140]gpt-5.1267/72025-11-13
58 [32–195]grok-4.20187/72026-03-10
59 [11–161]o3236/72025-04-16
60 [11–177]nemotron-3-ultra-550b-a55b4+13/72026-06-03
61 [8–228]nova-lite56/72024-12-03
62 [21–195]grok-4.5136/72026-07-08
63 [22–165]qwen2.5-32b-instruct45/72024-09-19
64 [20–176]gemini-3.5-flash207/72026-05-19
65 [18–183]internlm-chat-20b44/72023-09-18
66 [20–204]qwen3.5-9b3+15/72026-02-27
67 [29–157]gemini-1.5-pro207/72024-02-15
68 [40–146]kimi-k2.616+27/72026-04-14
69 [31–224]mistral-large66/72024-02-26
70 [3–184]claude-sonnet-5107/72026-06-30
71 [60–215]palmyra-fin63/72024-07-31
72 [44–157]kimi-k225+17/72025-07-11
73 [16–213]qwen3.6-max-preview33/72026-04-22
74 [20–156]o1166/72024-12-17
75 [30–178]qwen3-max66/72025-09-24
76 [48–149]gpt-oss-120b22+37/72025-08-05
77 [20–195]o1-mini117/72024-09-12
78 [21–198]palmyra-x563/72025-04-28
79 [27–195]grok-3-mini157/72025-04-03
80 [18–158]gemini-3-pro-preview227/72025-11-18
81 [45–164]deepseek-v3.222+17/72025-12-01
82 [39–190]gpt-oss-20b15+17/72025-08-05
83 [15–208]qwen3.7-max76/72026-05-20
84 [21–195]gpt-5.2-codex33/7
85 [46–189]glm-5.214+16/72026-06-16
86 [36–167]gemini-1.5-flash156/72024-05-14
87 [15–211]qwen3-coder-480b-a35b-instruct32/72025-07-22
88 [25–211]doubao-seed-1.645/72025-06-11
89 [52–186]grok-4.3197/72026-05-15
90 [43–195]gpt-5.3-codex55/72026-02-05
91 [49–172]kimi-k2.520+27/72026-01-01
92 [25–188]mimo-v2.55+15/72026-04-27
93 [45–196]grok-4.1-fast167/72025-11-19
94 [30–170]gemini-3.5-flash-lite6+15/72026-07-21
95 [38–174]qwen3-235b-a22b25+27/72025-04-29
96 [29–214]nemotron-3-super-120b-a12b7+16/72026-03-10
97 [45–186]minimax-m2.710+16/72026-04-09
98 [17–200]qwen3.6-plus66/72026-04-02
99 [52–175]gemini-2.5-pro337/72025-03-25
100 [44–161]grok-4277/72025-07-09
101 [45–174]claude-3-opus227/72024-03-04
102 [34–193]gpt-4147/72023-03-14
103 [19–220]qwen2.5-72b-instruct15+17/72024-09-16
104 [14–183]gpt-4.1-mini16+17/72025-04-14
105 [41–189]glm-5.114+26/72026-04-03
106 [36–178]gpt-oss-safeguard-20b3+15/72025-10-29
107 [33–162]gpt-4-turbo197/72023-11-06
108 [37–173]grok-4-fast125/72025-09-19
109 [13–234]phi-455/72024-12-11
110 [53–158]claude-3.5-sonnet277/72024-06-21
111 [25–246]phi-3-mini-4k-instruct43/72024-04-22
112 [38–221]deepseek-v3.110+17/72025-08-21
113 [38–231]deepseek-v4-pro127/72026-04-22
114 [48–182]qwen-14b-chat84/72023-09-25
115 [58–177]deepseek-v329+17/72024-12-26
116 [44–183]gemini-2.5-flash30+17/72025-04-17
117 [64–199]qwen3-next-80b-a3b106/72025-09-11
118 [12–222]qwen3.7-plus45/72026-05-26
119 [26–214]ernie-4.5-300b-a47b56/72025-06-28
120 [65–223]glm-4.656/72025-09-30
121 [47–209]gemma-4-26b-a4b-it4+15/72026-03-11
122 [42–196]gemma-4-31b-it7+17/72026-03-11
123 [26–232]minimax-m256/72025-10-27
124 [45–235]qwen3-30b-a3b-instruct46/72025-07-28
125 [59–182]qwen1.5-32b-chat46/72024-04-02
126 [58–181]baichuan-2-13b-chat125/72023-09-06
127 [59–254]mistral-medium45/72023-12-11
128 [34–265]qwen3-30b-a3b45/72025-04-27
129 [81–171]gpt-4o55+17/72024-05-13
130 [77–197]gemini-2-flash75/72024-12-11
131 [63–187]gemini-3.1-flash-lite18+27/72026-03-03
132 [28–251]glm-4.7-flash3+13/72026-01-19
133 [65–186]gemini-2-pro-preview63/72025-02-05
134 [61–199]qwen1.5-14b-chat65/72024-02-04
135 [63–185]gpt-3.5-turbo237/72023-03-01
136 [24–253]qwen3-32b5+17/72025-04-29
137 [80–189]deepseek-r1397/72025-01-20
138 [75–207]llama-3.3-70b-instruct22+27/72024-12-06
139 [12–224]granite-4.0-small-guardian63/72025-10-02
140 [61–180]qwen1.5-72b-chat126/72024-02-04
141 [76–177]o4-mini246/72025-04-16
142 [64–253]glm-4.765/72026-06-03
143 [65–205]gemini-3-flash-preview14+16/72025-12-17
144 [20–238]deepseek-v3.2-exp34/72025-09-29
145 [54–258]qwen3.5-122b-a10b35/72026-02-24
146 [66–203]llama-2-7b-chat127/72023-07-18
147 [22–221]nova-premier45/7
148 [66–193]yi-34b-chat85/72023-11-23
149 [24–242]nemotron-3-nano-30b-a3b4+14/72025-12-15
150 [68–222]minimax-m2.58+17/72026-02-12
151 [71–252]magistral-medium66/72025-06-10
152 [65–227]glm-4-9b-chat74/72024-06-05
153 [1–231]claude-2.132/72023-11-21
154 [59–227]text-davinci-00334/72022-11-28
155 [37–216]internlm-chat-7b54/72023-07-06
156 [54–225]mimo-v2-flash45/72025-12-16
157 [36–205]gemma-2b-it56/72024-02-21
158 [8–246]hy32+15/72026-07-02
159 [101–196]llama-4-maverick27+17/72025-04-05
160 [58–230]grok-3177/72025-04-03
161 [76–204]o3-mini236/72025-01-31
162 [54–213]llama-2-13b-chat86/72023-07-18
163 [56–213]granite-4.0-small63/72025-10-02
164 [65–200]gemini-2.0-flash126/72024-12-11
165 [97–201]qwen1.5-7b-chat96/72024-02-04
166 [58–206]command-r-plus107/72024-04-04
167 [58–215]gemma-7b-it77/72024-02-21
168 [65–193]claude-3-haiku177/72024-03-13
169 [61–217]gpt-5.4-mini146/72026-03-17
170 [91–202]llama-3.1-405b-instruct226/72024-07-23
171 [61–257]mistral-small-3.142/72025-03-11
172 [36–233]minimax-m2.143/72025-12-20
173 [75–224]glm-4.597/72025-07-28
174 [71–216]gemma-2-9b-it77/72024-06-27
175 [87–199]yi-6b-chat55/72023-11-23
176 [44–262]qwen3.5-35b-a3b34/72026-02-24
177 [79–192]gemini-2-flash-lite-preview63/72025-02-05
178 [100–231]gpt-4.1-nano126/72025-04-14
179 [29–251]hy3-preview43/72026-04-13
180 [106–218]gemma-2-27b-it57/72024-06-27
181 [65–235]llama-3.2-instruct-11b-vision34/7
182 [31–257]deepseek-llm-7b-chat33/72023-11-29
183 [44–258]command-r84/72024-03-11
184 [81–251]glm-5-turbo45/72026-03-15
185 [16–227]granite-4.0-micro-guardian63/72025-10-02
186 [91–215]llama-3-70b-instruct137/72024-04-18
187 [89–213]llama-3-8b-instruct157/72024-04-18
188 [82–238]gpt-5.4-nano125/72026-03-17
189 [111–212]qwen-7b-chat94/72023-08-03
190 [55–233]olmo-2-0325-32b-instruct76/72025-03-13
191 [53–213]llama-3.1-70b-instruct197/72024-07-23
192 [121–215]vicuna-13b33/72023-03-30
193 [53–227]gemini-2.5-flash-lite187/72025-06-17
194 [70–236]phi-3.5-mini-instruct56/72024-08-22
195 [131–217]baichuan-2-7b-chat115/72023-09-06
196 [107–237]chatglm-6b54/72023-03-14
197 [55–236]llama-2-70b-chat76/72023-07-18
198 [16–257]nvidia-nemotron-nano-9b-v232/72025-08-12
199 [144–205]claude-3-sonnet136/72024-03-04
200 [7–246]gemini-2.0-flash-lite44/72025-02-25
201 [78–252]qwen-2.5-7b-instruct15+17/72024-09-19
202 [133–222]chatglm3-6b94/72023-10-27
203 [72–255]mistral-small-3.235/72025-06-19
204 [107–222]qwen-2.5-14b-instruct65/72024-09-19
205 [98–250]step-3.5-flash3+15/72026-02-01
206 [79–261]phi-233/7
207 [20–247]yi-1.5-34b-chat34/72024-05-13
208 [46–243]nova-micro45/7
209 [157–221]gpt-4o-mini247/72024-07-18
210 [142–233]mistral-7b-instruct217/72023-09-27
211 [144–229]vicuna-7b-v1.544/72023-06-22
212 [119–249]mistral-medium-3.156/72025-08-12
213 [61–247]gemma-3-27b-it106/72025-03-12
214 [70–267]mistral-nemo45/72024-07-18
215 [84–263]mistral-medium-343/72025-05-07
216 [111–242]phi-4-mini34/7
217 [169–221]llama-4-scout167/72025-04-05
218 [27–254]granite-3.3-8b-instruct84/72025-04-10
219 [58–250]gemma-2-2b-it45/72024-07-31
220 [154–237]vicuna-13b-v1.544/72023-03-30
221 [130–248]qwen2.5-max46/72025-01-25
222 [131–247]mistral-small-444/72026-01-23
223 [148–240]llama-3.1-8b-instruct257/72024-07-23
224 [162–248]chatglm2-6b63/72023-06-25
225 [176–246]palmyra-med63/72024-07-31
226 [55–266]qwen2.5-3b-instruct33/72024-09-17
227 [76–263]qwen-2.5-1.5b-instruct53/72024-09-19
228 [137–249]llama-3.2-90b-vision-instruct33/72024-09-19
229 [25–260]olmo-2-1124-13b-instruct65/72024-11-26
230 [40–259]qwen1.5-0.5b-chat34/72024-02-04
231 [141–249]nova-pro66/72024-12-03
232 [72–237]granite-4.0-micro96/72025-10-02
233 [94–258]zephyr-7b-beta54/72023-10-26
234 [99–262]ling-2.6-flash3+15/7
235 [72–263]qwen2-7b-instruct43/72024-06-07
236 [174–246]vicuna-7b43/72023-06-22
237 [110–267]mistral-medium-3.534/72026-03-31
238 [179–266]mistral-small-386/72025-01-30
239 [150–257]qwen3-8b66/72025-04-29
240 [44–259]mixtral-8x22b-instruct95/72024-04-17
241 [175–256]moss-moon-003-sft33/72023-04-19
242 [176–253]trinity-large44/72026-01-27
243 [201–254]gemma-3-4b34/7
244 [193–255]command-a65/72025-03-11
245 [202–256]davinci34/72020-06-11
246 [189–262]qwen3-14b36/72025-04-27
247 [190–256]vicuna-33b-v1.344/72023-06-22
248 [158–261]mistral-large-396/72025-12-01
249 [135–265]deepseek-v3.2-speciale33/72025-11-28
250 [200–254]gemma-3-12b45/7
251 [113–268]deepseek-llm-67b-chat85/72023-11-29
252 [190–267]llama-3.2-1b-instruct76/72024-09-25
253 [198–259]llama-3.2-3b-instruct86/72024-09-25
254 [158–258]marin-8b-instruct53/72025-05-14
255 [125–266]magistral-small35/72025-06-04
256 [183–265]qwq-32b45/72025-03-06
257 [205–263]jamba-1-7-large34/7
258 [199–263]qwen-1.8b-chat33/72023-11-30
259 [198–263]olmo-2-1124-7b-instruct65/72024-11-26
260 [243–264]dbrx-instruct75/72024-03-27
261 [191–266]mixtral-8x7b-instruct106/72023-12-10
262 [169–268]grok-243/72024-12-12
263 [172–265]olmoe-1b-7b-0125-instruct53/72025-01-27
264 [227–264]mistral-large-2115/72024-07-24
265 [224–267]qwen1.5-1.8b-chat45/72024-02-04
266 [234–268]deepseek-r1-distill-llama-70b43/72025-01-20
267 [248–268]qwen-2.5-0.5b-instruct43/72024-09-19
268 [244–268]magistral-small-250934/72025-09-12