The Deterministic Horizon: When Extended Reasoning Fails and Tool Delegation Becomes Necessary

What it is — A paper identifying a “deterministic horizon” of 19–31 reasoning steps, past which models should delegate state-tracking to tools rather than keep reasoning internally.

Key points

• Tool-integrated approaches hit 86–94% accuracy versus 24–42% for pure chain-of-thought past that horizon.
• Fine-tuning on optimal reasoning traces only closed the gap by <5%, pointing to an architectural ceiling rather than a training problem.
• An “Attention Bottleneck Theorem” bounds state-tracking capacity; the effect held across 12 models and 8 task domains (r = 0.81–0.91).