diff --git a/.changeset/deepgram-flux-general-multi.md b/.changeset/deepgram-flux-general-multi.md
new file mode 100644
index 000000000..04a80b631
--- /dev/null
+++ b/.changeset/deepgram-flux-general-multi.md
@@ -0,0 +1,6 @@
+---
+'@livekit/agents-plugin-deepgram': patch
+'@livekit/agents': patch
+---
+
+Add Deepgram `flux-general-multi` STTv2 model support with multi-language detection. Introduces a new `languageHint` option for biasing the model toward specific languages (only used by `flux-general-multi`), and adds a new `sourceLanguages` field on `SpeechData` that carries all detected languages sorted by prevalence. For multi-language detection, the dominant language is set on `language` while `sourceLanguages` retains the full list.
diff --git a/agents/src/stt/stt.ts b/agents/src/stt/stt.ts
index 69ebd1418..48ceff5ba 100644
--- a/agents/src/stt/stt.ts
+++ b/agents/src/stt/stt.ts
@@ -64,6 +64,18 @@ export interface SpeechData {
   words?: TimedString[];
   /** Speaker identifier when the provider supports diarization. */
   speakerId?: string | null;
+  /**
+   * The source languages spoken by the user.
+   *
+   * Populated by STT services that support translation, where `language` holds the
+   * target language and `sourceLanguages` holds the original spoken language(s),
+   * or by multi-language detection services where `language` holds the dominant
+   * language and `sourceLanguages` holds all detected languages sorted by prevalence.
+   *
+   * May contain multiple entries when a single utterance spans multiple source languages.
+   */
+  // Ref: python livekit-agents/livekit/agents/stt/stt.py - 62-68 lines
+  sourceLanguages?: LanguageCode[];
 }
 
 export interface RecognitionUsage {
diff --git a/plugins/deepgram/src/models.ts b/plugins/deepgram/src/models.ts
index 11099893c..5a808965e 100644
--- a/plugins/deepgram/src/models.ts
+++ b/plugins/deepgram/src/models.ts
@@ -18,7 +18,8 @@ export type TTSModels =
 
 export type TTSEncoding = 'linear16' | 'mulaw' | 'alaw' | 'mp3' | 'opus' | 'flac' | 'aac';
 
-export type V2Models = 'flux-general-en';
+// Ref: python livekit-plugins/livekit-plugins-deepgram/livekit/plugins/deepgram/models.py - 38 line
+export type V2Models = 'flux-general-en' | 'flux-general-multi';
 
 export type STTModels =
   | 'nova-general'
diff --git a/plugins/deepgram/src/stt_v2.ts b/plugins/deepgram/src/stt_v2.ts
index 0c6c20dc4..2558bc787 100644
--- a/plugins/deepgram/src/stt_v2.ts
+++ b/plugins/deepgram/src/stt_v2.ts
@@ -36,6 +36,12 @@ export interface STTv2Options {
   eotTimeoutMs?: number;
   mipOptOut?: boolean;
   tags?: string[];
+  /**
+   * List of language hints to bias the model for improved accuracy.
+   * Only usable with `flux-general-multi`.
+   */
+  // Ref: python livekit-plugins/livekit-plugins-deepgram/livekit/plugins/deepgram/stt_v2.py - 61 line
+  languageHint?: string[];
 }
 
 const defaultSTTv2Options: Omit<STTv2Options, 'apiKey'> = {
@@ -104,6 +110,8 @@ export class STTv2 extends stt.STT {
    * @param opts.eotTimeoutMs - End-of-turn timeout in ms (default: 3000)
    * @param opts.keyterms - List of key terms to improve recognition
    * @param opts.tags - Tags for usage reporting (max 128 chars each)
+   * @param opts.languageHint - List of language hints to bias the model for improved accuracy.
+   *   Only usable with `flux-general-multi`.
    *
    * @throws Error if no API key is provided
    */
@@ -129,6 +137,18 @@ export class STTv2 extends stt.STT {
     if (this.#opts.tags) {
       this.#opts.tags = validateTags(this.#opts.tags);
     }
+
+    // Ref: python livekit-plugins/livekit-plugins-deepgram/livekit/plugins/deepgram/stt_v2.py - 134-138 lines
+    if (
+      this.#opts.languageHint &&
+      this.#opts.languageHint.length > 0 &&
+      this.#opts.model !== 'flux-general-multi'
+    ) {
+      this.#logger.warn(
+        { model: this.#opts.model },
+        '`languageHint` is only supported by `flux-general-multi` and will be ignored for this model',
+      );
+    }
   }
 
   /** The model being used for transcription */
@@ -165,8 +185,24 @@ export class STTv2 extends stt.STT {
    * @param opts - Partial options to update
    */
   updateOptions(opts: Partial<STTv2Options>) {
-    this.#opts = { ...this.#opts, ...opts };
+    this.#opts = {
+      ...this.#opts,
+      ...opts,
+      language:
+        opts.language !== undefined ? normalizeLanguage(opts.language) : this.#opts.language,
+    };
     if (opts.tags) this.#opts.tags = validateTags(opts.tags);
+    // Ref: python livekit-plugins/livekit-plugins-deepgram/livekit/plugins/deepgram/stt_v2.py - 244-249 lines
+    if (
+      this.#opts.languageHint &&
+      this.#opts.languageHint.length > 0 &&
+      this.#opts.model !== 'flux-general-multi'
+    ) {
+      this.#logger.warn(
+        { model: this.#opts.model },
+        '`languageHint` is only supported by `flux-general-multi` and will be ignored for this model',
+      );
+    }
     this.#logger.debug('Updated STTv2 options');
   }
 }
@@ -456,6 +492,11 @@ class SpeechStreamv2 extends stt.SpeechStream {
     if (this.#opts.keyterms.length > 0) params.keyterm = this.#opts.keyterms;
     if (this.#opts.tags && this.#opts.tags.length > 0) params.tag = this.#opts.tags;
 
+    // Ref: python livekit-plugins/livekit-plugins-deepgram/livekit/plugins/deepgram/stt_v2.py - 480-481 lines
+    if (this.#opts.languageHint && this.#opts.languageHint.length > 0) {
+      params.language_hint = this.#opts.languageHint;
+    }
+
     const baseUrl = this.#opts.endpointUrl.replace(/^http/, 'ws');
     const qs = queryString.stringify(params);
     return `${baseUrl}?${qs}`;
@@ -487,12 +528,20 @@ function parseTranscription(
     confidence = sum / wordsData.length;
   }
 
+  // Ref: python livekit-plugins/livekit-plugins-deepgram/livekit/plugins/deepgram/stt_v2.py - 587-591 lines
+  const detectedLanguagesRaw = Array.isArray(data.languages) ? (data.languages as string[]) : [];
+  const detectedLanguages = detectedLanguagesRaw.map((lang) => normalizeLanguage(lang));
+  const primaryLanguage =
+    detectedLanguages.length > 0 ? detectedLanguages[0]! : normalizeLanguage(language);
+
   const sd: stt.SpeechData = {
-    language: normalizeLanguage(language),
+    language: primaryLanguage,
     startTime: ((data.audio_window_start as number) || 0) + startTimeOffset,
     endTime: ((data.audio_window_end as number) || 0) + startTimeOffset,
     confidence: confidence,
     text: transcript || '',
+    // Ref: python livekit-plugins/livekit-plugins-deepgram/livekit/plugins/deepgram/stt_v2.py - 598 line
+    sourceLanguages: detectedLanguages.length > 0 ? detectedLanguages : undefined,
     // Note: Deepgram V2 (Flux) API does not provide word-level timing (start/end).
     // Words only contain 'word' and 'confidence' fields, so startTime/endTime will be 0.
     // See: https://developers.deepgram.com/docs/flux/nova-3-migration