Skip to main content

← Dracon Bench leaderboard

Head-to-head

nemotron-3-nano-omni-30b-a3b-reasoning vs nemotron-3-ultra-550b-a55b

The same frozen prompts at temperature 0, scored the same way. Every figure carries the items it was taken over, and the delta is always nemotron-3-nano-omni-30b-a3b-reasoning minus nemotron-3-ultra-550b-a55b.

Scores

nemotron-3-nano-omni-30b-a3b-reasoning against nemotron-3-ultra-550b-a55b.
Suitenemotron-3-nano-omni-30b-a3b-reasoningnemotron-3-ultra-550b-a55bDelta
Overall66.1% 6 of 6· #664.1% 6 of 6· #10+2.0 pts
IFEval-50 prompt88.0% n=50 76.2%–94.4%70.0% n=10 39.7%–89.2%+18.0 pts
IFBench-50 prompt57.1% n=49 43.3%–70.0%50.0% n=10 23.7%–76.3%+7.1 pts
MMLU-Pro-5032.0% n=50 20.8%–45.8%30.0% n=10 10.8%–60.3%+2.0 pts
JSON briefs100.0% n=5 56.6%–100.0%100.0% n=5 56.6%–100.0%tie
GSM8K-5078.0% n=50 64.8%–87.2%60.0% n=10 31.3%–83.2%+18.0 pts
SVG total104/250 blind rubric187/250 blind rubric-33.2 pts

Deltas compare scores, not denominators: a 10-point gap means less when one side answered far fewer items. The n= lines keep that honest.

beach

17 vs 33

Sun low over water with a reflection path; distant objects smaller and hazier than near ones.

nemotron-3-nano-omni-30b-a3b-reasoning

17/50

nemotron-3-ultra-550b-a55b

33/50

clock

31 vs 44

A clock is checkable to the minute: 10:10 means the hour hand just past 10, the minute hand on 2.

nemotron-3-nano-omni-30b-a3b-reasoning

31/50
12 1 2 3 4 5 6 7 8 9 10 11 DRACON

nemotron-3-ultra-550b-a55b

44/50
12 1 2 3 4 5 6 7 8 9 10 11 dracon

duo

20 vs 39

Two characters mid-action, not posed: a dog driving a jet ski, a raccoon holding on.

nemotron-3-nano-omni-30b-a3b-reasoning

20/50

nemotron-3-ultra-550b-a55b

39/50

monalisa

16 vs 31

Leonardo's portrait: three-quarter bust, folded hands, winding river landscape behind.

nemotron-3-nano-omni-30b-a3b-reasoning

16/50

nemotron-3-ultra-550b-a55b

31/50
after Leonardo

pelican

20 vs 40

Everyone knows what a pelican looks like: large beak with throat pouch, long neck, webbed feet.

nemotron-3-nano-omni-30b-a3b-reasoning

20/50

nemotron-3-ultra-550b-a55b

40/50