Raschka would spend a $100M LLM budget entirely on post-training

Asked how he'd build a SOTA LLM with $100M, Sebastian Raschka said he'd skip pretraining and spend it all on post-training an existing model. A reply claimed Jev reportedly used 100% synthetic data with no pretraining, yet still generalizes impressively.

2026-09-20 ~ 2026-09-20 · 3 related posts

• Jev reportedly trained on 100% synthetic data with no pretraining, just post-training — MaziyarPanahi · 2026-09-20
• Raschka: with $100M to build an LLM, I'd skip pretraining and invest in post-training — rasbt · 2026-09-20
• Raschka: with $100M for a top LLM, spend it all on post-training, not pretraining — MaziyarPanahi · 2026-09-20