← Evals

Evaluation profile

UGI Leaderboard — base-model willingness

2sub-evals
0%Safety weight
3.34%Freedom weight
1components

Weights below are portfolio-specific global index weights.

Model score (higher is better)Predicted score

About this eval

Willingness to answer and adhere to instructions on private sensitive-topic prompts.

Not included: not yet admitted by the scoring-source policy

Sub-evals

MeasureComponentDirection by rankingSafety weightFreedom weight
willingness_adherence_scoreugi-refusal/ugi-refusal.csv:willingness_adherence_scoreMeasures adherence rather than evasive deviation on sensitive requests; higher means more adherence.Freedom: user_direction_following:1.000ugi-refusalFreedom: higher1.67%
willingness_direct_scoreugi-refusal/ugi-refusal.csv:willingness_direct_scoreMeasures direct willingness to answer sensitive requests; higher means fewer direct refusals.Freedom: user_direction_following:1.000ugi-refusalFreedom: higher1.67%

willingness_adherence_score

Measures adherence rather than evasive deviation on sensitive requests; higher means more adherence.

RankModelValueRelative performanceProvenance
1qwen3-1.7b9official
2mistral-medium-38.5official
2qwen3-vl-4b-instruct8.5official
4mistral-large-38.25official
4qwen3-4b8.25official
6deepseek-v3.28official
6deepseek-v4-flash8official
6grok-38official
6mistral-small-38official
6mistral-small-3.18official
6mistral-small-3.28official
6qwen3-30b-a3b8official
13mistral-large-27.75official
13qwen3-8b7.75official
15grok-4.1-fast7.5official
15lfm2-8b-a1b7.5official
15mistral-medium-3.17.5official
18deepseek-v3.2-exp7.25official
19solar-10.7b-instruct7official
20magistral-small6.75official
21deepseek-v36.5official
21exaone-4.0-32b6.5official
21qwen3-vl-235b-a22b-instruct6.5official
24ministral-3-8b6.333official
25grok-46official
25internlm3-8b-instruct6official
25lfm2-24b-a2b6official
25llama-4-maverick6official
25qwen3-14b6official
25qwen3-vl-8b-instruct6official
25trinity-large6official
32ministral-3-14b5.667official
33deepseek-v3.1-terminus5.5official
33mixtral-8x7b-instruct5.5official
35grok-4-fast5.25official
35mistral-medium-3.55.25official
35mistral-small-45.25official
38mistral-nemo5official
38mistral-small-25official
38qwen3-0.6b5official
41deepseek-v3.14.75official
41glm-4.5-air4.75official
41qwen3.5-2b4.75official
44deepseek-v3.2-speciale4.5official
44grok-4.34.5official
44grok-4.54.5official
44muse-glimmer4.5official
44qwen2.5-14b-instruct4.5official
44qwen3-32b4.5official
50deepseek-v4-pro4.25official
51glm-4.7-flash4official
51qwen2.5-7b-instruct4official
53glm-4.6v3.75official
54mimo-v2-flash3.5official
54qwen3-235b-a22b3.5official
54reka-flash-33.5official
57claude-opus-4.63.375official
58glm-4.53.25official
58glm-4.63.25official
58mimo-v2.53.25official
61gemini-2.5-flash3official
61glm-4.73official
61gpt-oss-120b3official
61hy3-preview3official
61llama-3.3-70b-instruct3official
61minimax-m23official
67qwen3.5-0.8b2.75official
68deepseek-r12.5official
68qwq-32b2.5official
70gemma-4-31b-it2.25official
70nemotron-3-nano-30b-a3b2.25official
70qwen3.6-plus2.25official
73kimi-k22.167official
74kimi-k2.52official
74qwen2.5-1.5b-instruct2official
74qwen3-30b-a3b-instruct2official
74step-3.5-flash2official
78gpt-5.6-sol1.8official
79apriel-v1.6-15b-thinker1.5official
79claude-3-opus1.5official
79claude-fable-51.5official
79claude-opus-4.51.5official
79claude-opus-4.81.5official
79claude-sonnet-4.51.5official
79claude-sonnet-4.61.5official
79claude-sonnet-51.5official
79gemini-2.5-pro1.5official
79gemini-3-flash-preview1.5official
79gemini-3-pro-preview1.5official
79gemini-3.1-pro-preview1.5official
79gemini-3.5-flash1.5official
79gemma-4-26b-a4b-it1.5official
79glm-51.5official
79gpt-4.11.5official
79gpt-4o1.5official
79gpt-51.5official
79gpt-5.6-luna1.5official
79gpt-5.6-terra1.5official
79ling-1t1.5official
79mimo-v2.5-pro1.5official
79minimax-m2.11.5official
79o31.5official
79o4-mini1.5official
79olmo-3-32b-think1.5official
79qwen3-235b-a22b-instruct1.5official
79qwen3-4b-thinking1.5official
79qwen3-coder-30b-a3b-instruct1.5official
79qwen3-max1.5official
79qwen3-next-80b-a3b1.5official
79qwen3-next-80b-a3b-instruct1.5official
79qwen3-vl-235b-a22b1.5official
79qwen3-vl-30b-a3b1.5official
79qwen3-vl-32b-instruct1.5official
79qwen3.5-122b-a10b1.5official
79qwen3.5-27b1.5official
79qwen3.5-35b-a3b1.5official
79qwen3.5-397b-a17b1.5official
79qwen3.5-9b1.5official
79qwen3.6-27b1.5official
79ring-1t1.5official
79solar-pro-31.5official
122claude-haiku-4.51.25official
122kimi-k2.61.25official
122nvidia-nemotron-nano-9b-v21.25official
125claude-opus-4.71.125official
126gemma-2-27b-it1official
126gemma-2-9b-it1official
126gpt-oss-20b1official
126granite-4.0-small1official
126llama-3.2-1b-instruct1official
126minimax-m2.51official
126minimax-m2.71official
126nanbeige4.1-3b1official
126phi-41official
126qwen2.5-32b-instruct1official
126qwen3-4b-instruct1official
137claude-opus-40.75official
137claude-sonnet-40.75official
137glm-5.10.75official
137glm-5.20.75official
137qwen3.5-4b0.75official
137qwen3.6-35b-a3b0.75official
143seed-oss-36b-instruct0.6667official
144command-a0.5official
144llama-3.1-405b-instruct0.5official
144llama-3.2-3b-instruct0.5official
144olmo-3-7b-instruct0.5official
144olmo-3-7b-think0.5official
144qwen2.5-72b-instruct0.5official
144qwen3-vl-8b0.5official
151claude-3-haiku0official
151claude-opus-4.10official
151gemma-2-2b-it0official
151gemma-3-27b-it0official
151llama-3.1-70b-instruct0official
151llama-3.1-8b-instruct0official

willingness_direct_score

Measures direct willingness to answer sensitive requests; higher means fewer direct refusals.

RankModelValueRelative performanceProvenance
1grok-4.57official
1qwen3-vl-4b-instruct7official
3magistral-small6.5official
4grok-46official
4internlm3-8b-instruct6official
4lfm2-24b-a2b6official
4ling-1t6official
4mistral-medium-36official
4mistral-nemo6official
4qwen3-4b6official
4qwen3-4b-instruct6official
4qwen3-8b6official
4qwen3-max6official
4qwen3-vl-32b-instruct6official
4qwen3-vl-8b-instruct6official
4ring-1t6official
17deepseek-v35.5official
17glm-4.7-flash5.5official
17grok-4.1-fast5.5official
17mistral-medium-3.55.5official
21deepseek-v3.2-speciale5official
21grok-35official
21grok-4.35official
21llama-3.3-70b-instruct5official
21ministral-3-8b5official
21mistral-large-25official
21mistral-large-35official
21mistral-small-25official
21mistral-small-35official
21mistral-small-3.15official
21mistral-small-3.25official
21nanbeige4.1-3b5official
21phi-45official
21qwen2.5-32b-instruct5official
21qwen2.5-72b-instruct5official
21qwen3-1.7b5official
21qwen3-235b-a22b-instruct5official
21qwen3-vl-235b-a22b-instruct5official
21qwen3-vl-30b-a3b5official
21trinity-large5official
41deepseek-v3.24.5official
41deepseek-v4-flash4.5official
41qwen3-14b4.5official
44ministral-3-14b4.333official
44seed-oss-36b-instruct4.333official
46deepseek-v3.14official
46deepseek-v3.1-terminus4official
46exaone-4.0-32b4official
46gemini-2.5-flash4official
46gemma-3-27b-it4official
46glm-4.5-air4official
46granite-4.0-small4official
46lfm2-8b-a1b4official
46mimo-v2-flash4official
46mistral-medium-3.14official
46nvidia-nemotron-nano-9b-v24official
46olmo-3-7b-think4official
46qwen2.5-7b-instruct4official
46qwen3-30b-a3b-instruct4official
46qwen3-4b-thinking4official
46qwen3-next-80b-a3b-instruct4official
46qwen3-vl-8b4official
46reka-flash-34official
46solar-10.7b-instruct4official
65muse-glimmer3.75official
66deepseek-v3.2-exp3.5official
66glm-4.73.5official
66grok-4-fast3.5official
66mistral-small-43.5official
66nemotron-3-nano-30b-a3b3.5official
66qwen3-30b-a3b3.5official
66qwen3.5-0.8b3.5official
66qwen3.5-35b-a3b3.5official
66qwen3.6-27b3.5official
66qwen3.6-35b-a3b3.5official
76kimi-k23.333official
77gpt-5.6-luna3.2official
77gpt-5.6-sol3.2official
79apriel-v1.6-15b-thinker3official
79gemini-2.5-pro3official
79gemini-3.5-flash3official
79glm-4.63official
79glm-4.6v3official
79gpt-4.13official
79gpt-4o3official
79gpt-5.6-terra3official
79llama-3.1-70b-instruct3official
79mimo-v2.53official
79mimo-v2.5-pro3official
79o4-mini3official
79olmo-3-7b-instruct3official
79qwen2.5-1.5b-instruct3official
79qwen2.5-14b-instruct3official
79qwen3-0.6b3official
79qwen3-32b3official
79qwen3-coder-30b-a3b-instruct3official
79qwq-32b3official
79step-3.5-flash3official
99deepseek-r12.5official
99glm-5.22.5official
99hy3-preview2.5official
99kimi-k2.52.5official
99kimi-k2.62.5official
99qwen3.5-27b2.5official
99qwen3.5-2b2.5official
99qwen3.5-4b2.5official
99qwen3.5-9b2.5official
99qwen3.6-plus2.5official
109o32.333official
110claude-opus-4.62.25official
111claude-opus-4.82.2official
111claude-sonnet-52.2official
113claude-haiku-4.52official
113claude-opus-4.52official
113claude-opus-4.72official
113claude-sonnet-4.62official
113command-a2official
113deepseek-v4-pro2official
113gemini-3-flash-preview2official
113gemini-3-pro-preview2official
113gemini-3.1-pro-preview2official
113gemma-4-26b-a4b-it2official
113gpt-52official
113llama-3.1-405b-instruct2official
113llama-3.1-8b-instruct2official
113llama-3.2-1b-instruct2official
113llama-3.2-3b-instruct2official
113llama-4-maverick2official
113minimax-m22official
113minimax-m2.12official
113mixtral-8x7b-instruct2official
113olmo-3-32b-think2official
113qwen3-235b-a22b2official
113qwen3-vl-235b-a22b2official
113qwen3.5-397b-a17b2official
113solar-pro-32official
137claude-opus-41.5official
137claude-opus-4.11.5official
137claude-sonnet-41.5official
137claude-sonnet-4.51.5official
137gemma-4-31b-it1.5official
137glm-4.51.5official
137qwen3.5-122b-a10b1.5official
144gpt-oss-20b1.333official
145claude-fable-51.2official
146gemma-2-9b-it1official
146glm-5.11official
146gpt-oss-120b1official
146minimax-m2.51official
146minimax-m2.71official
146qwen3-next-80b-a3b1official
152glm-50.5official
153claude-3-haiku0official
153claude-3-opus0official
153gemma-2-27b-it0official
153gemma-2-2b-it0official