diff --git a/server/utils/AiProviders/ollama/index.js b/server/utils/AiProviders/ollama/index.js
index b88a8121870..7639c2e946a 100644
--- a/server/utils/AiProviders/ollama/index.js
+++ b/server/utils/AiProviders/ollama/index.js
@@ -263,6 +263,8 @@ class OllamaAILLM {
               prompt_tokens: res.prompt_eval_count,
               completion_tokens: res.eval_count,
               total_tokens: res.prompt_eval_count + res.eval_count,
+              // Override duration with Ollama's eval_duration for more accurate outputTps
+              duration: res.eval_duration / 1e9,
             },
           };
         })
@@ -282,8 +284,9 @@ class OllamaAILLM {
         prompt_tokens: result.output.usage.prompt_tokens,
         completion_tokens: result.output.usage.completion_tokens,
         total_tokens: result.output.usage.total_tokens,
-        outputTps: result.output.usage.completion_tokens / result.duration,
-        duration: result.duration,
+        outputTps:
+          result.output.usage.completion_tokens / result.output.usage.duration,
+        duration: result.output.usage.duration,
       },
     };
   }
@@ -349,6 +352,8 @@ class OllamaAILLM {
           if (chunk.done) {
             usage.prompt_tokens = chunk.prompt_eval_count;
             usage.completion_tokens = chunk.eval_count;
+            // Override duration with Ollama's eval_duration for more accurate outputTps
+            usage.duration = chunk.eval_duration / 1e9;
             writeResponseChunk(response, {
               uuid,
               sources,
diff --git a/server/utils/helpers/chat/LLMPerformanceMonitor.js b/server/utils/helpers/chat/LLMPerformanceMonitor.js
index bd02863edfe..f18a4278d9d 100644
--- a/server/utils/helpers/chat/LLMPerformanceMonitor.js
+++ b/server/utils/helpers/chat/LLMPerformanceMonitor.js
@@ -39,6 +39,8 @@ class LLMPerformanceMonitor {
   }
   /**
    * Wraps a function and logs the duration (in seconds) of the function call.
+   * If the output contains a `usage.duration` property, it will be used instead of the calculated duration.
+   * This allows providers to supply more accurate timing information.
    * @param {Function} func
    * @returns {Promise<{output: any, duration: number}>}
    */
@@ -47,7 +49,11 @@ class LLMPerformanceMonitor {
       const start = Date.now();
       const output = await func; // is a promise
       const end = Date.now();
-      return { output, duration: (end - start) / 1000 };
+      const calculatedDuration = (end - start) / 1000;
+      // Use duration from output.usage if provided (for providers that have more accurate timing)
+      // otherwise use the calculated duration from function start/end times
+      const duration = output?.usage?.duration ?? calculatedDuration;
+      return { output, duration };
     })();
   }
 
@@ -88,8 +94,13 @@ class LLMPerformanceMonitor {
 
       stream.metrics.total_tokens =
         stream.metrics.prompt_tokens + (stream.metrics.completion_tokens || 0);
-      stream.metrics.outputTps = stream.metrics.completion_tokens / duration;
-      stream.metrics.duration = duration;
+
+      // Use duration from reportedUsage if provided (for providers that have more accurate timing)
+      // otherwise use the calculated duration from stream start/end times
+      const effectiveDuration = reportedUsage.duration || duration;
+      stream.metrics.outputTps =
+        stream.metrics.completion_tokens / effectiveDuration;
+      stream.metrics.duration = effectiveDuration;
       return stream.metrics;
     };
     return stream;