We shipped a new version of the smart router on our v6 infrastructure, expanded the catalog with the newest cost-effective frontier models, and added a clearer place to read billed cost on every response.
- Router v6:
sansa-autonow runs on the latest matching stack, with routing decision time cut by about 5x - Sticky
session_id: reusing the samesession_idacross turns now caches the routing decision, bringing follow-up routing down to under 2ms and keeping the conversation on one model so provider prompt caches stay warm for larger cost savings than before - New models in the router: DeepSeek V4 Flash, DeepSeek V4 Pro, GLM 5.2, MiniMax M3, and Opus 5 are now eligible for smart routing
usage.cost: billed USD is returned onusage.costfor completions (including the streaming usage chunk) and on JSON transcription responses alongside token and duration fields
See Smart Routing for routing details and sticky sessions, and Models for the response shape and usage.cost.