diff --git a/src/agent/channel.rs b/src/agent/channel.rs
index 7bb716a59..44f4c2c8e 100644
--- a/src/agent/channel.rs
+++ b/src/agent/channel.rs
@@ -2491,7 +2491,15 @@ async fn transcribe_audio_attachment(
     }
 
     let (provider_id, model_name) = match deps.llm_manager.resolve_model(voice_model) {
-        Ok(parts) => parts,
+        Ok(parts) => {
+            tracing::debug!(
+                provider = %parts.0,
+                model = %parts.1,
+                voice_config = %voice_model,
+                "resolved voice model for transcription"
+            );
+            parts
+        }
         Err(error) => {
             tracing::warn!(%error, model = %voice_model, "invalid voice model route");
             return UserContent::text(format!(
@@ -2502,7 +2510,15 @@ async fn transcribe_audio_attachment(
     };
 
     let provider = match deps.llm_manager.get_provider(&provider_id) {
-        Ok(provider) => provider,
+        Ok(provider) => {
+            tracing::debug!(
+                provider = %provider_id,
+                base_url = %provider.base_url,
+                api_type = ?provider.api_type,
+                "got provider for voice transcription"
+            );
+            provider
+        }
         Err(error) => {
             tracing::warn!(%error, provider = %provider_id, "voice provider not configured");
             return UserContent::text(format!(
@@ -2527,6 +2543,11 @@ async fn transcribe_audio_attachment(
         "{}/v1/chat/completions",
         provider.base_url.trim_end_matches('/')
     );
+    tracing::debug!(
+        endpoint = %endpoint,
+        model = %model_name,
+        "sending voice transcription request"
+    );
     let body = serde_json::json!({
         "model": model_name,
         "messages": [{
diff --git a/src/llm/manager.rs b/src/llm/manager.rs
index 4febe761d..adc25dff5 100644
--- a/src/llm/manager.rs
+++ b/src/llm/manager.rs
@@ -289,12 +289,20 @@ impl LlmManager {
     }
 
     /// Resolve a model name to provider and model components.
-    /// Format: "provider/model-name" or just "model-name" (defaults to anthropic).
+    /// Format: "provider/model-name" or just "model-name" (defaults to openai for voice models).
     pub fn resolve_model(&self, model_name: &str) -> Result<(String, String)> {
         if let Some((provider, model)) = model_name.split_once('/') {
+            tracing::debug!(provider = %provider, model = %model, "resolved model with explicit provider");
             Ok((provider.to_string(), model.to_string()))
         } else {
-            Ok(("anthropic".into(), model_name.into()))
+            // Default to openai for voice models (most common for Whisper/vision)
+            // rather than anthropic, since anthropic doesn't support input_audio
+            tracing::debug!(
+                model = %model_name,
+                "no provider prefix specified in model name, defaulting to 'openai'. \
+                 Specify as 'provider/model' (e.g., 'openai/whisper-1') to use a different provider."
+            );
+            Ok(("openai".into(), model_name.into()))
         }
     }