Observed arrival · 2026-10-09
Sansar’s training log, down to the cloud bill
Sansar publishes a live ledger of model-training experiments across Sanskrit, English, math, OCR, and voice.
○Open
⊠Login
$Paid
†Ads
✦Pretty
●Pro
◎Niche
◉Human
⚑Risk
ƒJS
- For
- Model builders and Sanskrit-language researchers
- Worth noticing
- The dashboard reports 98 runs, 494 GPU hours, and $210 in cloud spend across five tracks.
Field notes
The run table exposes sort controls for date, model, headline score, decision, GPU hours, and cost. Its five tracks include experiments in Sanskrit, English, math, OCR, and voice, while dated milestones point to model releases and training updates. The voice section describes a goal of building separate voices for metrical verse and prose, but says audio from those experiments is not published.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
○OpenPublic substance visible
✦PrettyNotable craft visible
●ProPolished or operationally mature
◎NicheUnusually specific use
One card from the complete issue