Nvidia releases Nemotron-SFT-ARC-AGI-v1, agentic reasoning traces on ARC puzzles
GregKamradt · x · 2026-09-08
Nvidia published Nemotron-SFT-ARC-AGI-v1 on Hugging Face: 100K–1M multi-turn agentic reasoning traces from nine open-weight reasoning models solving ARC-AGI visual puzzles. Each puzzle is converted to text prompts; agents call a Python executor, an NL grid manipulator, and a solution-submission tool across turns. Eight subsets vary reasoning/tool availability; only successful runs are kept, intended for SFT of Nemotron 3 Ultra.
Related event: Nvidia Open-Sources ARC-AGI SFT Dataset as Google Builds Its Own(2 posts)→
More from Research
- Terence Tao: The Question Actually Had a Small Finite Counterexample — tak3sh8 · 2026-09-09
- Uno speeds up Qwen3-8B 2.5x by using diffusion for parallel token drafting — rohanpaul_ai · 2026-09-09
- 4.9M messages, 300B tokens, 10,000 agents: details of OpenAI's Navier–Stokes run — willdepue · 2026-09-09
- OpenAI's Navier-Stokes breakthrough: 10,000 agents solved it in 88 hours — generativist · 2026-09-09
- 16,893 sessions measured: what Claude Code, Codex and Cursor actually recommend — rseroter · 2026-09-09
- Largest open experiment yet on autonomous agents iterating in a research environment — eliebakouch · 2026-09-09