Introducing a new category of AI
POEM
The First Reductive Inference Model
88%
Accuracy
95.5×
Faster
100×
Smaller
Introducing a new category of AI
The First Reductive Inference Model
How it works
01
Every question arrives with a universe of wrong answers. POEM removes them first, before searching for what remains.
02
Each elimination step reduces the possibility space. The answer is not found — it is what cannot be eliminated.
03
What survives the elimination process is the answer. No generation. No guessing. Only what remains.
Benchmark results vs TinyLlama 1.1B
88% accuracy. 95.5× faster. 100× fewer parameters. Built on elimination, not generation.
10.9M parameters · 3 compute steps · 246,000 knowledge entries
Try POEM