Models · WFM (world foundation model, Cosmos3-Edge post-trained) · tic-tac-toe · vision-driven

agi-wfm/so101-ttt

checking availability…

Our checkpoint cagataydev/agi-wfm-so101-ttt (weights private for now), for the SO-101 arm. It answers one observation with the next 32 actions. It is not a general-purpose controller; the task sentence is passed to the model, but only the task it was trained on is expected to work. The older id wfm/so101-ttt still works and leases the same row.

tic-tac-toe · vision-driven.cell not yet selectable by language. The numbers below are held-out offline scores, not a success rate on a real arm — what the service can and cannot tell you.

Price

per call: $0.0032 · 32 tokens
per action: $0.0001
billed: every returned action, whether or not your robot plays it

What you send

observation.state [6]: shoulder_pan, shoulder_lift, elbow_flex, wrist_flex, wrist_roll, gripper · normalized: arm -100..100, gripper 0..100
observation.images.front [480, 640, 3]: front camera, RGB uint8 (JPEG on the wire)
observation.images.wrist [480, 640, 3]: wrist camera, RGB uint8 (JPEG on the wire)
task text: passed to the policy

Read the camera names out of your lease’s features (or the wire’s hello) rather than this table: a slot answers with its own.

What comes back

32 actions × 6 joints, to be played at 15 Hz — 2.13 s of motion per call.

Units before motion.Values are LeRobot normalized (arm −100…100, gripper 0…100), not degrees — whatever the semantics label says. That label echoes whichever runner answered your lease, so both values are in circulation: never branch on it (units, explained). LeRobot’s so101_follower defaults to degrees: run it with --robot.use_degrees=false. What the arm does if the connection drops.

Held out

ours: 14.17 chunk / 6.2 first step

14.17 over the 2.13 s it predicts (seed 0, 4 denoise steps, 263 held-out windows) against 19.80 for repeating the last action — 1.40x; across seeds it varies by 1.03. It picks and places from vision; the task string does not yet choose the cell..

RMSE over 2.133 s of motion (32 actions at 15 Hz) / first step, so101-normalized-v1 units — every gate below is over the same 2.133 s on the same 263 windows · cagataydev/so101-tic-tac-toe-norm (apache-2.0), the LAST episode of every task held out (19,29,…,209), every 15th frame · warm start from runs/20261025T092425Z-train (the flagship's svla post-training), 1,000 iterations ≈ 76k samples ≈ 1.9 epochs of 190 episodes, lr 3e-5, one H200, 181 min, $15.11. Lower is closer to the recorded human demonstration; it is not a task success rate.

The chunk here is its own 32 steps — 2.13 s of motion at 15 Hz. Do not read it against another model’s figure unless both the step count and the seconds of motion match: drift grows with time, so a chunk covering less of it is an easier average.

The deadline is not this model’s chunk. With agi rollout the arm asks for the next chunk when 0.75 s of motion is left in its queue (LeRobot’s own default is 0.5 s), so a round trip has roughly that long however long the chunk is — 2.13 s here only sets how often you pay for one. Answer later than that and the arm holds its last position targets until the chunk arrives; it does not stop. Your own round trip is the figure the dashboard and /play print after a call.

Try it

One billed call with synthetic inputs ($0.0032). No robot moves.

Code

pip install agi-diy
agi login
agi run agi-wfm/so101-ttt --probe -n 1

Needs Python 3.12 — LeRobot is installed on first use, requires 3.12 or later and does not run on 3.14 yet.

Or run a test call from the dashboard, or drive the simulated SO-101.

Licences and data

Checkpoint	cagataydev/agi-wfm-so101-ttt · weights private	OpenMDW-1.1 (base nvidia/Cosmos3-Edge)
Base model	nvidia/Cosmos3-Edge	OpenMDW-1.1
Training data	cagataydev/so101-tic-tac-toe-norm (apache-2.0), the LAST episode of every task held out (19,29,…,209), every 15th frame	as declared on the Hub

Licences as declared on the Hub. A lease says which of these answered it — pool and checkpoint on the session. Policies, explained.