feat(types): add presence_penalty and token breakdown details to LLM Gateway

CHANGELOG.md

@@ -1,5 +1,12 @@
 # Changelog
 
+## [4.41.6]
+
+- Expose `presence_penalty` in `ModelDefaultParameters` and `ChatCompletionRequest` — allows reading and configuring presence penalty for LLM Gateway models
+- Expose `thinking` in `ChatCompletionMessage` — surfaces reasoning/thinking content for reasoning models on the LLM Gateway
+- Expose detailed token breakdown fields (`prompt_tokens_details`, `completion_tokens_details`, `input_tokens`, `output_tokens`) in `ChatCompletionUsage` — surfaces cached tokens, cache creation, and reasoning token metrics
+- Expose optional `http_status_code`, `response_time`, `llm_status_code`, and `request` in `ChatCompletionResponse`
+
 ## [4.41.5]
 
 - Add `speakerLabelsRevisionIntervalMs` streaming param (sent as `speaker_labels_revision_interval_ms`) — the cadence, in ms of audio time, at which the server emits mid-stream `speakerRevision` events when `speakerLabels` is enabled. Unset or `0` sends only the end-of-stream revision; non-zero values are clamped server-side to 120 000–300 000 ms. The first mid-stream revision needs ~120 s of streamed speech, and one is sent only when some earlier turn's label changed

package.json

@@ -1,6 +1,6 @@
 {
   "name": "assemblyai",
-  "version": "4.41.5",
+  "version": "4.41.6",
   "description": "The AssemblyAI JavaScript SDK provides an easy-to-use interface for interacting with the AssemblyAI API, which supports async and real-time transcription.",
   "engines": {
     "node": ">=18"

src/types/llm-gateway/index.ts

@@ -4,6 +4,7 @@
 export type ChatCompletionMessage = {
   role: "user" | "assistant" | "system";
   content?: string | { type: string; text: string }[];
+  thinking?: string | null;
   tool_calls?: unknown[];
   tool_call_id?: string;
   name?: string;
@@ -21,6 +22,8 @@ export type ChatCompletionRequest = {
   max_tokens?: number;
   top_p?: number;
   top_k?: number;
+  frequency_penalty?: number;
+  presence_penalty?: number;
   tools?: unknown[];
   tool_choice?: unknown;
   response_format?: unknown;
@@ -29,15 +32,38 @@ export type ChatCompletionRequest = {
   [key: string]: unknown;
 };
 
+export type ChatCompletionCacheCreation = {
+  ephemeral_5m_input_tokens?: number | null;
+  ephemeral_1h_input_tokens?: number | null;
+};
+
+export type ChatCompletionPromptTokensDetails = {
+  cached_tokens?: number | null;
+  audio_tokens?: number | null;
+  cache_creation?: ChatCompletionCacheCreation | null;
+  cache_write_tokens?: number | null;
+};
+
+export type ChatCompletionTokensDetails = {
+  reasoning_tokens?: number | null;
+  audio_tokens?: number | null;
+  accepted_prediction_tokens?: number | null;
+  rejected_prediction_tokens?: number | null;
+};
+
 export type ChatCompletionUsage = {
   prompt_tokens: number;
   completion_tokens: number;
   total_tokens: number;
+  input_tokens?: number | null;
+  output_tokens?: number | null;
+  prompt_tokens_details?: ChatCompletionPromptTokensDetails | null;
+  completion_tokens_details?: ChatCompletionTokensDetails | null;
 };
 
 export type ChatCompletionChoice = {
   index: number;
-  finish_reason?: string;
+  finish_reason?: string | null;
   message: ChatCompletionMessage;
 };
 
@@ -48,12 +74,17 @@ export type ChatCompletionResponse = {
   request_id: string;
   choices: ChatCompletionChoice[];
   usage: ChatCompletionUsage;
+  http_status_code?: number | null;
+  response_time?: number | null;
+  llm_status_code?: number | null;
+  request?: unknown;
 };
 
 export type ModelDefaultParameters = {
   temperature?: number | null;
   top_p?: number | null;
   frequency_penalty?: number | null;
+  presence_penalty?: number | null;
 };
 
 export type ModelTopProvider = {

tests/unit/llm-gateway.test.ts

@@ -157,4 +157,114 @@ describe("llmGateway", () => {
       requestId: "req_000",
     });
   });
+
+  it("should parse detailed usage token breakdowns and thinking in chat completions", async () => {
+    const detailedResponse = {
+      request_id: "req_detailed_123",
+      http_status_code: 200,
+      response_time: 1542000000,
+      llm_status_code: 200,
+      choices: [
+        {
+          index: 0,
+          finish_reason: "stop",
+          message: {
+            role: "assistant",
+            content: "Here is the summary.",
+            thinking: "First analyze key points...",
+          },
+        },
+      ],
+      usage: {
+        input_tokens: 120,
+        output_tokens: 45,
+        prompt_tokens: 120,
+        completion_tokens: 45,
+        total_tokens: 165,
+        prompt_tokens_details: {
+          cached_tokens: 60,
+          audio_tokens: 0,
+          cache_creation: {
+            ephemeral_5m_input_tokens: 30,
+            ephemeral_1h_input_tokens: 0,
+          },
+          cache_write_tokens: 30,
+        },
+        completion_tokens_details: {
+          reasoning_tokens: 25,
+          audio_tokens: 0,
+          accepted_prediction_tokens: 0,
+          rejected_prediction_tokens: 0,
+        },
+      },
+    };
+    fetchMock.doMockOnceIf(
+      requestMatches({ url: "/v1/chat/completions", method: "POST" }),
+      JSON.stringify(detailedResponse),
+    );
+    const result = await assembly.llmGateway.chatCompletions({
+      model: "claude-haiku-4-5-20251001",
+      messages: [{ role: "user", content: "Summarize" }],
+      frequency_penalty: 0.5,
+      presence_penalty: 0.2,
+    });
+    expect(result.request_id).toBe("req_detailed_123");
+    expect(result.http_status_code).toBe(200);
+    expect(result.response_time).toBe(1542000000);
+    expect(result.choices[0].message.thinking).toBe(
+      "First analyze key points...",
+    );
+    expect(result.usage.prompt_tokens_details?.cached_tokens).toBe(60);
+    expect(
+      result.usage.prompt_tokens_details?.cache_creation
+        ?.ephemeral_5m_input_tokens,
+    ).toBe(30);
+    expect(result.usage.completion_tokens_details?.reasoning_tokens).toBe(25);
+    expect(requestBody()).toMatchObject({
+      frequency_penalty: 0.5,
+      presence_penalty: 0.2,
+    });
+  });
+
+  it("should parse presence_penalty in model default parameters", async () => {
+    const modelsWithPresencePenalty = {
+      data: [
+        {
+          id: "gpt-4o",
+          name: "GPT-4o",
+          description: "OpenAI GPT-4o",
+          default_parameters: {
+            temperature: 0.7,
+            top_p: 0.95,
+            frequency_penalty: 0.5,
+            presence_penalty: 0.2,
+          },
+          supported_parameters: [
+            "temperature",
+            "frequency_penalty",
+            "presence_penalty",
+          ],
+          top_provider: {
+            is_moderated: true,
+            context_length: 128000,
+            max_completion_tokens: 4096,
+          },
+          context_length: 128000,
+          pricing: { global: { completions: 10.0, prompt: 2.5 } },
+          creator: "openai",
+          retirement_date: 0,
+          available_regions: ["global"],
+          providers: ["openai"],
+          default_provider: "openai",
+        },
+      ],
+    };
+    fetchMock.doMockOnceIf(
+      requestMatches({ url: "/v1/models", method: "GET" }),
+      JSON.stringify(modelsWithPresencePenalty),
+    );
+    const result = await assembly.llmGateway.listModels();
+    expect(result.data[0].default_parameters.presence_penalty).toBe(0.2);
+    expect(result.data[0].default_parameters.frequency_penalty).toBe(0.5);
+  });
 });