TEMPO Training Boosts ARC-AGI-3 Scores by Over 30%

dots3-note's TEMPO mechanism recursively segments multi-hour trajectories into macro-steps for process rewards, beating baselines by 31.5% on ARC-AGI-3 and GRPO by 20.6%, while its test-time learning lets it self-correct and reuse experience in unseen environments.

2026-08-30 ~ 2026-08-31 · 3 related posts

• TEMPO: recursive self-critique splits day-long agent rollouts into macro-steps — SarahAnnabels · 2026-08-30
• dots3-note shows test-time learning: explores, self-corrects, reuses knowledge in unseen environments — SarahAnnabels · 2026-08-30
• TEMPO Outperforms Baselines on ARC-AGI-3 with Self-Correction — CodeByPoonam · 2026-08-31