Compute is where AI discovery quietly gets expensive. Screen everything, and you pay for every structure that never had a chance. MS Lattice spends only where the proof points.
The numbers
- ≈ $0.003of compute per candidate in the blind benchmark
- ≈ 30 sof compute per candidate
- ≈ 38 sper compound, to rank the right structure from a formula
- < $0.003for an agent’s entire 8-candidate campaign
In the blind benchmark
64 candidates, 32 minutes, under 20 cents, both sealed targets found.
At scale
At about a third of a cent of compute per candidate: 1,000 candidates for about $3. 10,000 for about $30. 100,000 for about $300. You see the price before you pay it.
Why it is so efficient
MS Lattice learns where to look while it runs, and nothing is spent on candidates that could never qualify.









