From dd3bcd34213ed7a3731be251e1ee9403781123c6 Mon Sep 17 00:00:00 2001
From: Praveen11558 <44603409+Praveen11558@users.noreply.github.com>
Date: Thu, 5 Mar 2026 12:28:06 +0530
Subject: [PATCH] token_count calculation during inline data sent during
 multimodal inputs

---
 litellm/litellm_core_utils/llm_cost_calc/utils.py | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/litellm/litellm_core_utils/llm_cost_calc/utils.py b/litellm/litellm_core_utils/llm_cost_calc/utils.py
index e7c09f6ec0c..5809b8c4d8c 100644
--- a/litellm/litellm_core_utils/llm_cost_calc/utils.py
+++ b/litellm/litellm_core_utils/llm_cost_calc/utils.py
@@ -663,7 +663,7 @@ def generic_cost_per_token(  # noqa: PLR0915
     accounted_tokens = text_tokens + cache_hit + audio_tokens + cache_creation + image_tokens
     has_double_counting = cache_hit > 0 and accounted_tokens > usage.prompt_tokens
      # Double-counting fix (xAI etc.): recalculate text_tokens from scratch
-    if has_double_counting:
+    if (text_tokens == 0 and prompt_tokens_details["image_count"] == 0) or has_double_counting:
         text_tokens = (
             usage.prompt_tokens
             - cache_hit