Codex Cache Misses Fall in T3 Code

This is a meaningful infrastructure improvement, though the cross-harness comparison is directional rather than a controlled benchmark.

• –Sol’s miss rate dropped roughly 29% month over month.
• –The gap with Opus narrowed from 1.7 percentage points to 0.7.
• –Better cache reuse can reduce both latency and inference costs for long coding sessions.
• –Rolling-window metrics avoid stale baselines but may make short-term trends more volatile.
• –T3 Code is emerging as a useful real-world observability layer for comparing coding-agent performance.