feat: expose service_tier in CompletionUsage from OpenAI Responses API

OpenAI returns service_tier (e.g. "default", "priority", "flex") in
every API response. This adds it to CompletionUsage and reads it in
the Responses plugin's handleResponseCompleted.

Changes:
- Add serviceTier?: string to CompletionUsage interface
- Read event.response.service_tier in Responses LLM

Author: piyush-gambhir

.changeset/expose-service-tier.md

@@ -0,0 +1,6 @@
+---
+'@livekit/agents': patch
+'@livekit/agents-plugin-openai': patch
+---
+
+expose serviceTier in CompletionUsage from OpenAI Responses API

agents/src/llm/llm.ts

@@ -25,6 +25,8 @@ export interface CompletionUsage {
   promptTokens: number;
   promptCachedTokens: number;
   totalTokens: number;
+  /** The service tier used for processing (e.g. 'default', 'priority', 'flex'). */
+  serviceTier?: string;
 }
 
 export interface ChatChunk {

plugins/openai/src/responses/llm.ts

@@ -333,6 +333,7 @@ class ResponsesHttpLLMStream extends llm.LLMStream {
           promptTokens: event.response.usage.input_tokens,
           promptCachedTokens: event.response.usage.input_tokens_details.cached_tokens,
           totalTokens: event.response.usage.total_tokens,
+          serviceTier: event.response.service_tier ?? undefined,
         },
       };
     }

plugins/openai/src/ws/llm.ts

@@ -584,6 +584,7 @@ export class WSLLMStream extends llm.LLMStream {
           promptTokens: event.response.usage.input_tokens,
           promptCachedTokens: event.response.usage.input_tokens_details.cached_tokens,
           totalTokens: event.response.usage.total_tokens,
+          serviceTier: (event.response as Record<string, unknown>).service_tier as string | undefined,
         },
       };
     }