Reasoning v2 engine

The reasoning engine powering Deep Mode has been rebuilt. Reasoning v2 replaces the fixed-depth search strategy in v1 with an adaptive approach that allocates reasoning steps based on problem complexity.

Simple tasks resolve faster. Hard tasks get more depth where it counts.

Performance gains on crucible-1:

• DOCLENS-v2 score: 91.4 (up from 77.8)

• Median Deep Mode latency: down 22%

• Reasoning trace length on simple tasks: down 31%

The upgrade is automatic. No changes to your integration required.