Local Model Benchmarks

Local test machine: Apple M5 Pro · 64 GB unified memoryRun window 29 September — 18 October 202622 recorded runs

Explore by task

Ten agentic challenges and one inference suite. Compare attempts at the same prompt, or inspect a single recorded result.

Same task, different setups

Local builds and cloud baselines

Whiteout Protocol TensorFold scene with the weapon, snowy arena and pause prompt visible

Whiteout Protocol

2 local runs · 1 cloud baseline

Wave survival, first-person combat and a snowy arena. Two local configurations and a cloud baseline attempted the same prompt.

Digger 3D voxel maze with the player, enemies and level HUD visible

Digger 3D

1 local run · 1 cloud baseline

Deterministic grid simulation, excavation and five levels in an isometric Three.js scene.

Hardball 3D TensorFold gameplay from behind the batter, with the inning scoreboard and field visible

Hardball 3D

2 local runs · 1 cloud baseline

Batting, pitching, fielding and a full inning. Compare the generated games and their recorded checks.

Obby 3D gameplay at stage one, showing the player, floating platforms and stage HUD

Obby 3D

2 local runs

A ten-stage third-person obstacle course with desktop and touch controls, checkpoints and varied hazards.

Long briefs, run by the task-loop orchestrator

Fresh sessions, no human input

Command-line app

add · import · report · budget · undo

Node.js, no dependencies

Pocket Ledger

1 local run

A dependency-free command-line finance tracker, planned and built over many fresh sessions with no human input.

Other experiments