From 16d3d28138008d001eeb0d64e57af352b263ddae Mon Sep 17 00:00:00 2001
From: octo-patch <266937838+octo-patch@users.noreply.github.com>
Date: Wed, 29 Jul 2026 15:34:43 +0000
Subject: [PATCH] Enable MiniMax models and wire their image/video request
 format

Add minimax/MiniMax-M3 and minimax/MiniMax-M2.7 to the allowed model
list and format MiniMax multimodal messages using the OpenAI-compatible
image_url/video_url schema so MiniMax-M3 can be used for image and video
evaluation instead of raising on multimodal input.
---
 mllm_tools/litellm.py         | 31 ++++++++++++++++++++++++++++++-
 src/utils/allowed_models.json |  4 +++-
 2 files changed, 33 insertions(+), 2 deletions(-)

diff --git a/mllm_tools/litellm.py b/mllm_tools/litellm.py
index 4873494ad..a266b7f3b 100644
--- a/mllm_tools/litellm.py
+++ b/mllm_tools/litellm.py
@@ -148,8 +148,37 @@ def __call__(self, messages: List[Dict[str, Any]], metadata: Optional[Dict[str,
                         })
                     else:
                         raise ValueError("For GPT, only text and image inferencing are supported")
+                elif "minimax" in self.model_name.lower():
+                    # MiniMax uses the OpenAI-compatible content schema and
+                    # supports both image and video inputs.
+                    if msg["type"] == "image":
+                        formatted_messages.append({
+                            "role": "user",
+                            "content": [
+                                {
+                                    "type": "image_url",
+                                    "image_url": {
+                                        "url": data_url
+                                    }
+                                }
+                            ]
+                        })
+                    elif msg["type"] == "video":
+                        formatted_messages.append({
+                            "role": "user",
+                            "content": [
+                                {
+                                    "type": "video_url",
+                                    "video_url": {
+                                        "url": data_url
+                                    }
+                                }
+                            ]
+                        })
+                    else:
+                        raise ValueError("For MiniMax, only text, image and video inferencing are supported")
                 else:
-                    raise ValueError("Only support Gemini and Gpt for Multimodal capability now")
+                    raise ValueError("Only support Gemini, Gpt and MiniMax for Multimodal capability now")
 
         try:
             # if it's openai o series model, set temperature to None and reasoning_effort to "medium"
diff --git a/src/utils/allowed_models.json b/src/utils/allowed_models.json
index 2ca44ee76..35071367a 100644
--- a/src/utils/allowed_models.json
+++ b/src/utils/allowed_models.json
@@ -13,6 +13,8 @@
     "bedrock/anthropic.claude-3-5-sonnet-20240620-v1:0",
     "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0",
     "bedrock/anthropic.claude-3-5-haiku-20241022-v1:0",
-    "bedrock/us.anthropic.claude-3-7-sonnet-20250219-v1:0"
+    "bedrock/us.anthropic.claude-3-7-sonnet-20250219-v1:0",
+    "minimax/MiniMax-M3",
+    "minimax/MiniMax-M2.7"
   ]
 }
\ No newline at end of file