Nvidia’s announcement framed Nemotron 3 Super — 120B total parameters, 12B active, a hybrid SSM latent-MoE design built for Blackwell — around the claim that it sees no wall in post-training as scaling RL keeps paying off (post). The release extends the Nemotron 3 family that Nvidia open-sourced in December with weights, data and training infrastructure rather than weights alone. In the trajectory it is the supply-side companion to the same issue’s PostTrainBench v1.0: one says post-training has room to run, the other measures whether agents can run it themselves, and Altman’s thousandfold cost drop in the same paragraph is the price curve underneath both.
Development
Nvidia sees no wall in post-training and opens a 120B model
Nvidia's no-wall reading of post-training, delivered with a fully open release, is the supply-side counterpart to a benchmark measuring whether agents can run that post-training themselves.
Claim
Nvidia said it sees no wall in post-training and announced Nemotron 3 Super, a 120-billion-parameter hybrid model with 12 billion active parameters, released with permissive licensing, open data and open training infrastructure.
Reported metric: 120B / 12B active
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to March 12, 2026 (theinnermostloop.substack.com)
- No wall in post-training: Nemotron 3 Super announced (x.com)
Verification record
These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.
- Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.