Physics World Models for the Engineered World

KYAMOS

Holy Grail of Multiphysics Simulations

Powered by TUNet-AI, our proprietary artificial intelligence for fast, physics-aware engineering predictions.

Non-autoregressive: Predicts directly without chaining earlier predictions.
21,900×: Faster than the reference numerical calculation in the selected test.
99.8%: Agreement with the analytical solution (exact mathematical answer) in the selected test.

Company-reported performance figures from a selected shock-tube test (sudden pressure change). Results depend on the application. See the test context and evidence.

Time & space aware
Geometry & material aware
Boundary, initial & source aware
Multiphysics aware
Scale aware · dimensionless numbers
Reference comparisons

Explore Solutions Discover TUNet-AI

Member of NVIDIA Inception

Anti-UAV410 · Thermal drone tracking

UNOFFICIAL WORLD RECORD

Company-reported; independent verification pending

Concept animation

Selected successful tracking example87.8% box overlap in this selected frame only; not overall tracking accuracy. Actual saved prediction from the 7 June 2026 research snapshot, with an enlarged target view.

KYAMOS reports surpassing the published best result used for comparison on this thermal-video drone-tracking benchmark.

4%

Absolute score gain

The reported gain means four percentage points on a 0–100 score scale. For example, 70 to 74 is a four-point gain; it does not mean a 4% relative increase.

Company-reported, unpublished result. The exact model version, comparison scores and matching test conditions remain pending. Read the drone-tracking result context.

See anti-UAV results and pictures

The saved champion combines baseline and gap-filled predictions with a validation-selected router, rather than a standalone TUNet model. This selected frame and the June snapshot do not establish the ranking claim above. Earlier failure comparisons remain in the research details.

Explore what we build