diff --git a/src/models.test.ts b/src/models.test.ts
index a91ed8b1..c63310dd 100644
--- a/src/models.test.ts
+++ b/src/models.test.ts
@@ -277,6 +277,15 @@ describe("capability flags vs blockrun's catalog", () => {
       expect(BLOCKRUN_MODELS.find((m) => m.id === id)?.vision).toBe(true);
     }
   });
+
+  it("marks MiniMax M3 vision-capable so image_url requests reach it", () => {
+    // The target metadata lists text+image+video input for MiniMax M3, but the
+    // model was registered without `vision`, so filterByVision() excluded it
+    // whenever a request carried an image_url content part. M2.7 is text-only
+    // upstream and must stay without the flag.
+    expect(BLOCKRUN_MODELS.find((m) => m.id === "minimax/minimax-m3")?.vision).toBe(true);
+    expect(BLOCKRUN_MODELS.find((m) => m.id === "minimax/minimax-m2.7")?.vision).toBeUndefined();
+  });
 });
 
 describe("OPENCLAW_MODELS integrity", () => {
diff --git a/src/models.ts b/src/models.ts
index 5333d31e..0f102cad 100644
--- a/src/models.ts
+++ b/src/models.ts
@@ -1358,6 +1358,7 @@ export const BLOCKRUN_MODELS: BlockRunModel[] = [
     contextWindow: 1048576,
     maxOutput: 65536,
     reasoning: true,
+    vision: true,
     agentic: true,
     toolCalling: true,
   },