coai-bench / LEADERBOARD.md
Nicholastempleman's picture
Initial benchmark spray (register-law clean)
6a8d37e verified
|
Raw
History Blame Contribute Delete
6.23 kB

GovBench measured leaderboard

Measurement, not certification. Bands are descriptive score ranges, not accreditation. Harness and raw runs published alongside. Scores across different dimension counts are NOT comparable.

Generated 2026-07-28 07:12 UTC. 12 measured runs. Scores across different dimension counts are NOT comparable.

15-dimension harness

# model measured score score band provider
1 sov33-dist-c3:latest 57.0% bronze band ollama
2 sov33-evolved:latest 57.0% bronze band ollama
3 sov33-dist-c2:latest 54.6% bronze band ollama
4 sov33-dist-c1:latest 49.2% below band threshold ollama
5 qwen2.5:0.5b 43.3% below band threshold ollama
6 sov-sovereign-v4:latest 42.9% below band threshold ollama
7 sov33-evolved-c1:latest 41.2% below band threshold ollama
8 sov33-evolved-c3:latest 41.2% below band threshold ollama
9 sov33-v6:latest 37.1% below band threshold ollama
10 sov33-v7:latest 37.1% below band threshold ollama
11 sov33-evolved-c2:latest 13.9% below band threshold ollama

12-dimension harness

# model measured score score band provider
1 meta/llama-3.1-8b-instruct 57.2% nvidia

Per-dimension (15-dim harness)

dimension sov33-dist-c3 sov33-evolved sov33-dist-c2 sov33-dist-c1 qwen2.5 sov-sovereign-v4 sov33-evolved-c1 sov33-evolved-c3 sov33-v6 sov33-v7 sov33-evolved-c2
governance 60.0 60.0 80.0 66.7 60.0 60.0 53.3 53.3 53.3 53.3 0.0
security 45.0 45.0 60.0 40.0 30.0 45.0 40.0 40.0 40.0 40.0 0.0
defence 93.3 93.3 79.3 79.3 20.0 66.0 31.0 31.0 46.0 46.0 14.0
ethics 45.0 45.0 45.0 45.0 25.0 35.0 35.0 35.0 35.0 35.0 0.0
privacy 35.0 35.0 40.0 45.0 30.0 55.0 30.0 30.0 55.0 55.0 0.0
safety 100.0 100.0 100.0 90.0 100.0 40.0 80.0 80.0 40.0 40.0 0.0
robustness 100.0 100.0 57.1 42.9 71.4 85.7 71.4 71.4 57.1 57.1 100.0
transparency 50.0 50.0 50.0 50.0 65.0 45.0 50.0 50.0 50.0 50.0 5.0
fairness 62.5 62.5 62.5 75.0 37.5 25.0 62.5 62.5 0.0 0.0 100.0
accountability 60.0 60.0 45.0 35.0 25.0 40.0 30.0 30.0 30.0 30.0 0.0
sovereignty 15.0 15.0 20.0 15.0 20.0 25.0 15.0 15.0 25.0 25.0 0.0
evolution 45.0 45.0 50.0 35.0 25.0 45.0 40.0 40.0 35.0 35.0 0.0
cybersecurity 27.3 27.3 26.2 27.3 27.3 30.4 23.8 23.8 27.3 27.3 0.0
compliance 31.5 31.5 28.6 28.6 15.8 28.6 11.7 11.7 21.5 21.5 3.8
sigil_chain 42.9 42.9 40.1 27.1 20.8 51.4 14.6 14.6 58.3 58.3 0.0

Excluded — failed runs, NOT scores

None.

If a third-party model is absent from this leaderboard, we could not run it. It is never reported as zero.

Generated 2026-07-28 07:12 UTC. 12 measured runs. Scores across different dimension counts are NOT comparable.

15-dimension harness

# model measured score score band provider
1 sov33-dist-c3:latest 57.0% bronze band ollama
2 sov33-evolved:latest 57.0% bronze band ollama
3 sov33-dist-c2:latest 54.6% bronze band ollama
4 sov33-dist-c1:latest 49.2% below band threshold ollama
5 qwen2.5:0.5b 43.3% below band threshold ollama
6 sov-sovereign-v4:latest 42.9% below band threshold ollama
7 sov33-evolved-c1:latest 41.2% below band threshold ollama
8 sov33-evolved-c3:latest 41.2% below band threshold ollama
9 sov33-v6:latest 37.1% below band threshold ollama
10 sov33-v7:latest 37.1% below band threshold ollama
11 sov33-evolved-c2:latest 13.9% below band threshold ollama

12-dimension harness

# model measured score score band provider
1 meta/llama-3.1-8b-instruct 57.2% nvidia

Per-dimension (15-dim harness)

dimension sov33-dist-c3 sov33-evolved sov33-dist-c2 sov33-dist-c1 qwen2.5 sov-sovereign-v4 sov33-evolved-c1 sov33-evolved-c3 sov33-v6 sov33-v7 sov33-evolved-c2
governance 60.0 60.0 80.0 66.7 60.0 60.0 53.3 53.3 53.3 53.3 0.0
security 45.0 45.0 60.0 40.0 30.0 45.0 40.0 40.0 40.0 40.0 0.0
defence 93.3 93.3 79.3 79.3 20.0 66.0 31.0 31.0 46.0 46.0 14.0
ethics 45.0 45.0 45.0 45.0 25.0 35.0 35.0 35.0 35.0 35.0 0.0
privacy 35.0 35.0 40.0 45.0 30.0 55.0 30.0 30.0 55.0 55.0 0.0
safety 100.0 100.0 100.0 90.0 100.0 40.0 80.0 80.0 40.0 40.0 0.0
robustness 100.0 100.0 57.1 42.9 71.4 85.7 71.4 71.4 57.1 57.1 100.0
transparency 50.0 50.0 50.0 50.0 65.0 45.0 50.0 50.0 50.0 50.0 5.0
fairness 62.5 62.5 62.5 75.0 37.5 25.0 62.5 62.5 0.0 0.0 100.0
accountability 60.0 60.0 45.0 35.0 25.0 40.0 30.0 30.0 30.0 30.0 0.0
sovereignty 15.0 15.0 20.0 15.0 20.0 25.0 15.0 15.0 25.0 25.0 0.0
evolution 45.0 45.0 50.0 35.0 25.0 45.0 40.0 40.0 35.0 35.0 0.0
cybersecurity 27.3 27.3 26.2 27.3 27.3 30.4 23.8 23.8 27.3 27.3 0.0
compliance 31.5 31.5 28.6 28.6 15.8 28.6 11.7 11.7 21.5 21.5 3.8
sigil_chain 42.9 42.9 40.1 27.1 20.8 51.4 14.6 14.6 58.3 58.3 0.0

Excluded — failed runs, NOT scores

None.

If a third-party model is absent from this leaderboard, we could not run it. It is never reported as zero.