Which language models have the strongest safety evidence?

An evidence-weighted ranking of 925 language models across 161 independent evaluations. 346 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 166 ranks.

Rank by release date

346 of 346 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. Gemini 4 Argon Google · 1 evals across 1 components
  2. GPT-6.1 Sol OpenAI · 5 evals across 4 components
  3. Claude Sonnet 5.5 Anthropic · 5 evals across 4 components
  4. Space Bunny Alpha Stealth · 3 evals across 6 components
  5. Solar Mini 4 Upstage · 4 evals across 6 components
  6. GPT-6 Sol OpenAI · 13 evals across 7 components
  7. GPT-6 Luna OpenAI · 12 evals across 7 components
  8. Claude Opus 5.5 Anthropic · 11 evals across 7 components

Full ranking table

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

346 models
Overall rank; brackets show the 90% source-sensitivity interval.
Estimated rank
[90% interval]
ModelEvalsComponentsRelease date
1 [1–5]gpt-6.1-sol54/72026-09-29
2 [1–8]gpt-6-astra187/72026-09-03
3 [4–22]claude-opus-5277/72026-07-24
4 [3–15]gpt-6-sol10+37/72026-09-22
5 [7–37]claude-fable-5277/72026-06-09
6 [3–75]claude-opus-5.5117/72026-09-22
7 [8–50]claude-opus-4.8337/72026-05-28
8 [3–72]gpt-6-luna9+37/72026-09-22
9 [4–94]claude-fable-5.1207/72026-09-01
10 [4–19]muse-spark-1.35+15/72026-09-02
11 [7–69]muse-spark-1.1177/72026-07-09
12 [17–72]gpt-5.6-terra267/72026-07-09
13 [6–89]gpt-5.3-chat65/72026-03-03
14 [7–87]gpt-5.5-instant53/72026-05-26
15 [2–82]claude-sonnet-5.554/72026-09-28
16 [4–133]claude-mythos-preview33/72026-04-07
17 [18–67]muse-spark-1.285/72026-08-05
18 [15–76]claude-opus-4.7387/72026-04-16
19 [17–81]claude-opus-4.6377/72026-02-05
20 [24–76]gpt-5.5457/72026-04-23
21 [23–77]gpt-5.4387/72026-03-05
22 [13–137]inkling-small2+35/72026-07-30
23 [17–67]grok-4.78+17/72026-09-21
24 [12–102]gemini-3.7-flash196/72026-08-13
25 [15–88]gemini-3.8-flash18+17/72026-09-02
26 [24–127]gpt-5.6-sol357/72026-07-09
27 [9–167]granite-4.2-8b1+25/72026-08-25
28 [10–165]mimo-v2-omni33/72026-03-18
29 [23–77]deepseek-v4.1-flash4+23/72026-09-10
30 [7–114]gpt-5.2-chat54/72025-12-11
31 [16–121]glm-5.3-flash6+16/72026-08-26
32 [17–103]claude-sonnet-4.6397/72026-01-21
33 [31–88]glm-5.315+16/72026-08-18
34 [22–151]gpt-5.6-luna25+37/72026-07-09
35 [29–91]gpt-5.2357/72025-12-11
36 [16–194]claude-mythos-537/72026-06-09
37 [17–130]gemini-3.6-flash116/72026-07-21
38 [18–169]gpt-5-nano25+27/72025-08-07
39 [5–169]claude-295/72023-07-11
40 [34–100]claude-haiku-4.5477/72025-10-15
41 [3–267]gpt-5.5-pro33/72026-04-23
42 [19–124]claude-opus-4267/72025-05-22
43 [34–116]kimi-k3267/72026-07-16
44 [37–119]gpt-5-mini32+17/72025-08-07
45 [4–278]gpt-5-pro54/72025-08-07
46 [4–195]gemini-2.5-pro-exp43/72025-03-25
47 [19–129]claude-opus-4.5297/72025-11-24
48 [9–202]phi-3.5-moe-instruct56/72024-08-22
49 [20–253]qwen3.5-27b76/72026-02-24
50 [4–194]o1-preview46/72024-09-12
51 [18–156]gpt-5457/72025-08-07
52 [12–159]grok-4.20-multi-agent44/72026-03-10
53 [49–113]qwen3.5-397b-a17b12+17/72026-02-16
54 [15–122]claude-3.7-sonnet287/72025-02-24
55 [37–131]claude-sonnet-4.5447/72025-09-29
56 [39–137]gpt-5.1347/72025-11-13
57 [15–161]gpt-5.1-codex32/72025-11-13
58 [9–312]granite-4.0-h-small34/72025-10-02
59 [1–283]gpt-5.4-pro43/72026-03-05
60 [48–136]claude-sonnet-4407/72025-05-22
61 [22–188]solar-mini41+36/72026-09-23
62 [21–221]claude-sonnet-5227/72026-06-30
63 [11–239]llama-3.1-tulu-3-8b35/72024-11-20
64 [55–139]claude-opus-4.1216/72025-08-05
65 [27–186]mimo-v2.6-pro4+37/72026-09-21
66 [61–142]grok-4.5217/72026-07-08
67 [8–209]mimo-v2-pro55/72026-03-18
68 [22–214]mimo-v2.5-pro8+16/72026-04-27
69 [48–146]gemini-3.1-pro-preview347/72026-02-19
70 [30–211]qwen3.5-9b4+17/72026-02-27
71 [46–185]gemini-3.1-pro76/72026-02-19
72 [53–159]gpt-oss-20b19+17/72025-08-05
73 [66–170]minimax-m39+26/72026-06-02
74 [18–246]inkling8+26/72026-07-14
75 [32–224]gpt-5.4-mini236/72026-03-17
76 [13–256]nemotron-3.5-lightning2+15/72026-08-11
77 [16–205]claude-3.5-haiku157/72024-10-22
78 [62–149]kimi-k2.622+27/72026-04-14
79 [41–334]palmyra-x-00463/72024-10-09
80 [56–214]gemini-3.5-flash307/72026-05-19
81 [59–176]qwen3.5-plus76/72026-02-15
82 [33–259]glm-4.5-air13+15/72025-07-28
83 [13–289]nova-lite76/72024-12-03
84 [37–196]gemma-4-26b-a4b-it7+15/72026-03-11
85 [7–284]chatgpt-4o44/72025-03-27
86 [46–181]gpt-4.1377/72025-04-14
87 [29–230]qwen3.7-max147/72026-05-20
88 [39–187]gemini-1.5-pro227/72024-02-15
89 [30–319]muse-spark53/72026-04-08
90 [109–286]palmyra-fin63/72024-07-31
91 [29–195]o1196/72024-12-17
92 [27–247]qwen3-4b45/72025-04-27
93 [38–291]mistral-large66/72024-02-26
94 [30–279]qwen3.7-flash2+12/72026-07-27
95 [48–332]qwen3-235b-a22b-fp8-throughput63/72025-04-28
96 [29–231]gemini-1.0-pro86/72023-12-06
97 [68–228]glm-515+17/72026-02-11
98 [32–295]internlm2-chat-7b44/72024-01-17
99 [73–201]gemma-4-31b-it15+27/72026-03-11
100 [44–339]ministral-3-3b2+13/72025-10-31
101 [66–189]qwen3-max96/72025-09-24
102 [23–254]palmyra-x563/72025-04-28
103 [34–189]qwen2-72b-instruct105/72024-06-07
104 [34–201]gpt-4.5-preview116/72025-02-27
105 [55–211]laguna-s-2.1-poolside2+36/72026-07-21
106 [49–265]deepseek-v3.114+17/72025-08-21
107 [81–199]kimi-k232+17/72025-07-11
108 [76–228]minimax-m2.715+17/72026-04-09
109 [24–204]o3306/72025-04-16
110 [61–248]mimo-v2.56+25/72026-04-27
111 [77–246]deepseek-v4-flash19+17/72026-04-22
112 [30–252]o1-mini117/72024-09-12
113 [49–211]gemini-1.5-flash177/72024-05-14
114 [83–186]gpt-oss-120b32+37/72025-08-05
115 [36–221]gpt-oss-safeguard-20b3+15/72025-10-29
116 [48–250]qwen-72b-chat44/72023-11-30
117 [43–273]seed-2.0-mini2+15/72026-02-14
118 [37–244]internlm-chat-20b44/72023-09-18
119 [24–278]deepseek-v3.2-exp75/72025-09-29
120 [55–252]gpt-5.2-codex33/72025-12-18
121 [78–238]glm-5.225+17/72026-06-16
122 [21–278]qwen2.5-72b-instruct18+17/72024-09-16
123 [75–293]gpt-5.3-codex55/72026-02-05
124 [20–271]hy34+25/72026-07-02
125 [84–196]gemini-3-pro-preview277/72025-11-18
126 [23–264]qwen3.6-27b45/72026-04-21
127 [78–191]gemini-2.5-pro477/72025-03-25
128 [44–238]gemini-3.5-flash-lite8+15/72026-07-21
129 [63–255]longcat-2.02+15/72026-07-05
130 [74–286]minimax-m276/72025-10-27
131 [83–220]grok-4.3247/72026-05-15
132 [41–288]internlm2-chat-20b55/72024-01-17
133 [47–227]qwen3-235b-a22b30+27/72025-04-29
134 [28–224]gpt-4.1-mini21+17/72025-04-14
135 [93–214]deepseek-v3.232+27/72025-12-01
136 [59–320]olmo-3-7b-instruct33/72025-11-19
137 [92–238]doubao-seed-1.666/72025-06-11
138 [27–316]nemotron-3-nano-omni-30b-a3b2+15/72026-04-28
139 [36–332]qwen3-30b-a3b65/72025-04-27
140 [25–250]phi-4-reasoning-plus35/72025-04-17
141 [67–245]grok-3-mini177/72025-04-03
142 [82–228]claude-3-opus277/72024-03-04
143 [91–250]grok-4-fast186/72025-09-19
144 [26–261]ernie-4.5-300b-a47b66/72025-06-28
145 [10–319]nova-2.0-lite42/72025-12-02
146 [14–274]ox-alpha3+25/72026-08-20
147 [23–305]qwen3-32b8+17/72025-04-29
148 [68–197]claude-3.5-sonnet347/72024-06-21
149 [73–273]kimi-k2.7-code44/72026-06-11
150 [71–267]glm-5.118+27/72026-04-03
151 [87–273]minimax-m2.512+17/72026-02-12
152 [83–204]gpt-4-turbo257/72023-11-06
153 [1–316]gemma-4-e4b34/72026-03-31
154 [81–299]olmo-3.1-32b-instruct32/72025-12-10
155 [40–310]phi-3-mini-4k-instruct43/72024-04-22
156 [73–227]deepseek-v333+17/72024-12-26
157 [36–290]qwen3.8-max96/72026-08-03
158 [15–287]phi-465/72024-12-11
159 [96–269]grok-4.20277/72026-03-10
160 [78–230]gemini-2.5-flash40+17/72025-04-17
161 [78–249]qwen-14b-chat84/72023-09-25
162 [103–226]grok-4337/72025-07-09
163 [35–266]space-bunny-alpha36/72026-09-23
164 [100–275]qwen3.6-plus117/72026-04-02
165 [12–284]granite-4.0-small-guardian63/72025-10-02
166 [64–274]qwen2.5-32b-instruct55/72024-09-19
167 [123–205]gpt-4o72+17/72024-05-13
168 [76–294]deepseek-v4-pro237/72026-04-22
169 [77–295]qwen3.6-max-preview53/72026-04-22
170 [122–277]qwen3-next-80b-a3b126/72025-09-11
171 [108–215]gemini-2.0-pro-preview63/72025-02-05
172 [43–337]qwen3-max-preview32/72025-09-05
173 [87–280]qwen3.7-plus85/72026-05-26
174 [109–265]kimi-k2.530+27/72026-01-01
175 [52–288]grok-4.6156/72026-08-12
176 [43–326]mercury-23+14/72026-02-24
177 [81–277]grok-3197/72025-04-03
178 [14–301]qwen3.8-27b2+23/72026-08-14
179 [84–331]mistral-medium45/72023-12-11
180 [78–251]gpt-4187/72023-03-14
181 [56–273]nemotron-3-ultra-550b-a55b9+17/72026-06-03
182 [97–237]qwen1.5-32b-chat46/72024-04-02
183 [33–280]nova-premier45/72025-04-30
184 [124–295]qwen3-235b-a22b-instruct34/72025-07-21
185 [46–329]glm-4.7-flash5+13/72026-01-19
186 [102–239]baichuan-2-13b-chat135/72023-09-06
187 [23–292]nova-micro65/72024-12-03
188 [99–242]claude-3-haiku197/72024-03-13
189 [124–240]o3-mini266/72025-01-31
190 [93–303]qwen3.5-flash5+15/72026-02-24
191 [47–314]deepseek-v4-flash-07314+13/72026-07-31
192 [121–268]llama-3.3-70b-instruct30+27/72024-12-06
193 [107–279]nemotron-3-super-120b-a12b10+16/72026-03-10
194 [102–258]qwen1.5-14b-chat65/72024-02-04
195 [82–258]text-davinci-00334/72022-11-28
196 [42–317]nemotron-3-nano-30b-a3b7+14/72025-12-15
197 [92–242]gpt-3.5-turbo287/72023-03-01
198 [69–288]command-r-plus127/72024-04-04
199 [138–265]deepseek-r1487/72025-01-20
200 [123–240]gemini-3.1-flash-lite25+27/72026-03-03
201 [123–258]gemini-2.0-flash207/72024-12-11
202 [112–247]qwen1.5-72b-chat136/72024-02-04
203 [137–244]o4-mini346/72025-04-16
204 [151–265]llama-4-maverick35+17/72025-04-05
205 [109–288]glm-4-9b-chat84/72024-06-05
206 [93–279]qwen3-30b-a3b-instruct76/72025-07-28
207 [66–290]solar-pro-42+25/72026-08-11
208 [113–253]yi-34b-chat95/72023-11-23
209 [99–272]granite-4.0-small63/72025-10-02
210 [119–311]llama-3.2-instruct-11b-vision34/72024-09-25
211 [20–288]gpt-5.4-nano166/72026-03-17
212 [134–271]llama-3.1-405b-instruct236/72024-07-23
213 [42–312]hy3-preview55/72026-04-13
214 [123–342]mistral-small-3.152/72025-03-11
215 [75–268]gemma-2b-it56/72024-02-21
216 [59–291]internlm-chat-7b54/72023-07-06
217 [115–290]gpt-4.1-nano156/72025-04-14
218 [48–325]deepseek-llm-7b-chat33/72023-11-29
219 [125–268]gemini-3-flash-preview28+17/72025-12-17
220 [141–257]qwen1.5-7b-chat96/72024-02-04
221 [119–276]llama-2-13b-chat106/72023-07-18
222 [92–273]gemma-7b-it77/72024-02-21
223 [91–303]step-3.5-flash5+16/72026-02-01
224 [27–323]qwen3-coder-480b-a35b-instruct42/72025-07-22
225 [16–284]granite-4.0-micro-guardian63/72025-10-02
226 [89–278]claude-2.154/72023-11-21
227 [112–314]grok-4.1-fast257/72025-11-19
228 [43–308]gemini-2.0-flash-lite55/72025-02-25
229 [139–268]llama-2-7b-chat167/72023-07-18
230 [113–269]gemma-2-9b-it97/72024-06-27
231 [48–320]mistral-medium-3.575/72026-03-31
232 [35–317]llama-3.2-90b-vision-instruct44/72024-09-19
233 [142–298]glm-4.5117/72025-07-28
234 [130–314]glm-4.796/72026-06-03
235 [132–268]yi-6b-chat55/72023-11-23
236 [127–295]mistral-small-476/72026-01-23
237 [131–265]claude-3-sonnet157/72024-03-04
238 [64–306]step-3.7-flash1+22/72026-05-23
239 [131–272]llama-3-70b-instruct137/72024-04-18
240 [131–290]mimo-v2-flash76/72025-12-16
241 [129–303]glm-4.696/72025-09-30
242 [156–261]llama-3-8b-instruct167/72024-04-18
243 [109–282]gemini-2.5-flash-lite237/72025-06-17
244 [145–323]magistral-medium86/72025-06-10
245 [109–308]deepseek-v3.1-terminus44/72025-09-22
246 [52–307]qwen3.5-4b35/72026-02-27
247 [141–307]gemini-2.0-flash-lite-preview75/72025-02-05
248 [67–333]command-r84/72024-03-11
249 [88–313]gemma-3-12b96/72025-03-10
250 [46–309]mimo-v2.6-flash4+27/72026-09-21
251 [115–320]qwen3.5-122b-a10b55/72026-02-24
252 [102–304]yi-1.5-9b-chat33/72024-05-13
253 [58–312]glm-5-turbo76/72026-03-15
254 [171–278]gemma-2-27b-it67/72024-06-27
255 [59–300]olmo-2-32b-instruct86/72025-03-13
256 [157–278]llama-3.1-70b-instruct237/72024-07-23
257 [81–304]qwen2.5-7b-instruct19+17/72024-09-19
258 [118–301]mercury-2.51+23/72026-09-08
259 [127–290]chatglm-6b54/72023-03-14
260 [104–304]phi-3.5-mini-instruct56/72024-08-22
261 [160–281]qwen-7b-chat94/72023-08-03
262 [140–323]qwen3.5-35b-a3b54/72026-02-24
263 [73–316]ernie-5.134/72026-05-09
264 [183–277]baichuan-2-7b-chat115/72023-09-06
265 [115–335]phi-233/72023-12-12
266 [187–282]gpt-4o-mini317/72024-07-18
267 [26–333]nvidia-nemotron-nano-9b-v232/72025-08-12
268 [116–301]llama-2-70b-chat96/72023-07-18
269 [99–306]gemma-3-27b-it136/72025-03-10
270 [58–305]yi-1.5-34b-chat44/72024-05-13
271 [97–325]o3-pro43/72025-06-10
272 [103–313]nova-pro86/72024-12-03
273 [200–293]chatglm3-6b104/72023-10-27
274 [150–287]qwen2.5-14b-instruct65/72024-09-19
275 [81–326]mistral-medium-3.197/72025-08-12
276 [198–292]vicuna-7b-v1.544/72023-06-22
277 [105–313]minimax-m2.163/72025-12-20
278 [198–287]llama-4-scout217/72025-04-05
279 [48–333]mistral-small-396/72025-01-30
280 [111–306]trinity-large74/72026-01-27
281 [218–308]mistral-7b-instruct247/72023-09-27
282 [110–324]mistral-small-3.245/72025-06-19
283 [167–338]qwen3-vl-32b-instruct33/72025-10-19
284 [107–321]gemma-2-2b-it45/72024-07-31
285 [216–314]palmyra-med63/72024-07-31
286 [202–303]vicuna-13b-v1.544/72023-03-30
287 [120–332]olmo-3.1-32b-think32/72025-12-10
288 [176–327]qwen2-7b-instruct54/72024-06-07
289 [82–328]granite-3.3-8b-instruct94/72025-04-10
290 [182–330]qwen2.5-max46/72025-01-25
291 [139–343]mistral-nemo55/72024-07-18
292 [190–325]vicuna-13b43/72023-03-30
293 [141–337]qwen2.5-1.5b-instruct53/72024-09-19
294 [80–343]qwen2.5-3b-instruct33/72024-09-17
295 [165–321]muse-glimmer2+14/72026-08-10
296 [203–326]chatglm2-6b63/72023-06-25
297 [87–328]qwen1.5-0.5b-chat34/72024-02-04
298 [207–324]gemma-3-4b66/72025-03-10
299 [141–340]ling-2.6-flash3+15/72026-04-21
300 [31–332]olmo-2-13b-instruct65/72024-11-26
301 [209–314]qwen3-14b66/72025-04-27
302 [226–314]llama-3.1-8b-instruct327/72024-07-23
303 [165–330]zephyr-7b-beta54/72023-10-26
304 [47–339]mixtral-8x22b-instruct95/72024-04-17
305 [132–342]seallms-v3-7b-chat34/72024-07-03
306 [178–344]mistral-medium-365/72025-05-07
307 [243–316]qwen3-8b106/72025-04-29
308 [168–340]mistral-7b44/72023-09-27
309 [241–315]command-a85/72025-03-11
310 [159–332]phi-4-mini44/72025-02-26
311 [238–328]moss-moon-003-sft33/72023-04-19
312 [247–322]qwen2.5-7b33/72024-09-19
313 [124–343]qwen3-next-80b-a3b-instruct42/72025-09-09
314 [234–335]llama-3.2-1b-instruct96/72024-09-25
315 [142–333]granite-4.0-micro96/72025-10-02
316 [252–330]vicuna-33b-v1.344/72023-06-22
317 [185–339]deepseek-v3.2-speciale33/72025-11-28
318 [240–331]davinci34/72020-06-11
319 [142–340]qwen3-vl-235b-a22b-instruct44/72025-09-22
320 [246–331]vicuna-7b53/72023-06-22
321 [184–335]mistral-large-3166/72025-12-01
322 [224–343]qwen3-vl-8b-instruct44/72025-10-11
323 [209–339]grok-build-0.134/72026-05-20
324 [153–346]deepseek-llm-67b-chat85/72023-11-29
325 [208–334]marin-8b-instruct53/72025-05-14
326 [220–339]qwen3-4b-instruct32/72025-04-29
327 [245–341]mixtral-8x7b-instruct136/72023-12-10
328 [234–341]qwq-32b45/72025-03-06
329 [283–333]jamba-1.7-large44/72025-07-03
330 [217–339]olmo-2-7b-instruct65/72024-11-26
331 [297–336]jamba-1.7-mini34/72025-07-03
332 [283–336]mistral-large-2145/72024-07-24
333 [297–340]command-r7b34/72024-12-13
334 [257–341]qwen-1.8b-chat33/72023-11-30
335 [305–339]dbrx-instruct75/72024-03-27
336 [279–346]llama-3.2-3b34/72024-09-25
337 [268–341]olmoe-1b-7b-instruct65/72025-01-27
338 [295–343]qwen1.5-1.8b-chat45/72024-02-04
339 [230–346]grok-243/72024-12-12
340 [293–343]deepseek-r1-distill-llama-70b43/72025-01-20
341 [271–344]llama-3.2-3b-instruct85/72024-09-25
342 [295–345]magistral-small55/72025-06-04
343 [318–346]qwen2.5-0.5b-instruct43/72024-09-19
344 [284–346]llama-3.1-8b-base34/72024-07-23
345 [320–345]ministral-3-14b32/72025-10-31
346 [333–346]ministral-3-8b42/72025-10-31