@posthog/ai@8.9.2
Immutable
release. Only release title and notes can be modified.
8.9.2
Patch Changes
- #4664
86fe9daThanks @bernatixer! - Interrupted or cancelled streams now report the token usage and latency they actually observed, instead of zeros, across the OpenAI, Anthropic, Gemini, Azure and Vercel wrappers. When usage was never reported, token counts and override costs are omitted entirely, so$ai_input_tokenscan be absent where it was previously always0.
(2026-09-01)