Session-level observability can make the relationship between provider routing, prompt caching, and spend inspectable, even though it does not itself prove savings.
The mechanism is: Grouping turns into sessions exposes provider continuity and cache behavior alongside cost; affinity keeps successive turns on the preceding provider, where more prompt content may be cached. The log view then presents a session’s duration, cache hit rate, and turn-by-turn path. Multi-turn coding and chat agents make individual request logs a poor unit for understanding provider continuity and accumulated cost.
The source confirms the feature’s described fields and behavior, but provides no independent performance measurements, before-and-after cost data, or evidence that affinity improves outcomes for every workload. The evidence boundary is: The source is a product changelog describing fields and behavior; it offers no independent performance measurements, before-and-after cost data, or evidence that affinity improves every workload. Teams using AI Gateway can inspect session-level routing and cache behavior without reconstructing conversations from scattered requests. Whether teams can show lower per-session cost or latency while preserving routing reliability. This is a distinct analytical record because no prior New Runtime record is supplied for this specific session-logging release.