Post-training Nemotron with Era data for $609 matches frontier models at 88.9%, 27x cheaper per correct answer
NVIDIAAI · x · 2026-10-17
Era's team got early access to NVIDIA's Nemotron 3.5 Super VL and post-trained it with impressive results: the model matched frontier models at 88.9% accuracy with 27x lower cost per correct answer using Spot GPUs. They used Era-generated data to post-train Nemotron for just $609, improving its ability to recognize missing information while further cutting inference costs. The most exciting part, they say, is closing the loop between benchmarking and post-training: find where a model fails, generate data to fix it, and validate the new version against an entirely different simulated company.
More from Models
- AI scientist slams Anthropic: Claude anthropomorphization is dangerous, secrecy betrays science — rbhar90 · 2026-10-17
- Hash Rate Ep. 189: Jev models on Bittensor subnet SN92, decentralized training vs traditional LLMs — markjeffrey · 2026-10-17
- 'Remove AI tics and mannered prose': a one-line prompt fix for Argon's writing — arjunrajlab · 2026-10-17
- Blind test of 200 real image edits: nano banana pro loses to its cheaper base model — ravisparikh · 2026-10-17
- Dev ditches Gemini 4 Argon for Grok 4.5, which eats 96.6% of his Cursor tokens — gaganghotra_ · 2026-10-17
- LMArena opens Code Arena WebDev for blind head-to-head model coding battles — arena · 2026-10-17