Notes from theSituated World Models
Notes from MachineFi Labs on Situated World Models, Trio-Spark v1.0, and the research behind intelligence built for specific environments.
Trio-Spark v1.4: one model for text and vision, same API
Trio-Spark now runs one model for text, image and sampled-video decisions. Thirteen benchmarks against Jev 1.14, Clef, Clef-Flash and Laya.
Read the post→
Trio-Spark v1.3: Better decisions on every task we measured
New text weights for the same one-pass decision API. Thirteen benchmarks against v1.2, v1.0 and four other fast decision models.
Read the post→
Trio-Spark v1.2: Sharper decisions. Same fast API.
A stronger text release for the same decision API. Matched results, separate offline comparisons, and two to eight choices your application controls.
Read the post→
Trio-Spark v1.1: See the scene. Choose the next move.
Bring text, one image, or a short sampled window to the same decision API—and choose from the actions your application allows.
Read the post→
Introducing Trio-Spark v1.0: A one-pass decision API for GUI agents
A structured state and 2–8 allowed actions in; one selected action and the full probability distribution out.
Read the post→Trio: A World Model for Physical Operations
A status update from MachineFi Labs on Trio — what we've built, what's running now, and how to get your cameras on it.
Read the post→