No benchmark numbers published yet — the locked test set needs more held-out launches with observed outcomes.
What this version relies on
held-out validation launches
This version was trained before Model Zero started measuring which signals each model relies on.
For each signal, how much the score drops when that signal is scrambled across launches. Model Zero learns these weights itself from the data; nobody hand-codes them.
Training runs
Attempt 1Failed
14 Oct 15:53 · 00:47:48 · — · $0.00
Reason: training process terminated before finishing (no final status written)
Attempt 2Running
14 Oct 16:50 · — · — · cost not reported
Timeline
14 Oct 16:50
Model Zero v0.68 restarted
It had stalled (a failed run (retrying)); the watchdog restarted that step.
14 Oct 16:50
Training started: Model Zero v0.68 (attempt 2)
Solana Launch Dataset v44 · boosted decision trees · Model Zero training server (CPU)
14 Oct 16:50
Retrying Model Zero v0.68
Attempt 2. The failed attempt stays in the history.
14 Oct 16:41
Model Zero v0.68 training FAILED
Reason: training process terminated before finishing (no final status written). Cost: $0.00. Awaiting a controlled retry.
14 Oct 15:53
Training started: Model Zero v0.68
Solana Launch Dataset v44 · boosted decision trees · Model Zero training server (CPU)
14 Oct 15:52
Model Zero v0.68 queued
Automatic search from v0.66: min samples leaf 40 → 10; min samples leaf 10 → 5, with automatic signal discovery.
14 Oct 15:52
Enough new data for another training run
277,191 new labelled snapshots since Solana Launch Dataset v9.