From c363f65f19a341b99d6dbfb492480e00ba2a4bec Mon Sep 17 00:00:00 2001
From: "jinjing.zzj" <jinjing.zzj@alibaba-inc.com>
Date: Thu, 27 Aug 2026 18:35:50 +0800
Subject: [PATCH] fix(core): recognize new DeepSeek/GLM vision models in
 modality auto-detection

Add targeted entries to MODALITY_PATTERNS so name-based modality
detection grants image input to the vision models reported in #10270:

- deepseek-v4-flash-vision-exp (and any DeepSeek -vision variant)
  previously matched the family-wide [/^deepseek/, {}] text-only entry.
- glm-4.6v / glm-5v-turbo belong to the same v-suffix vision family as
  the already-recognized glm-4.5v; generalize that entry to
  [/^glm-[0-9.]+v/, { image: true }].
- glm-5.3-flash natively integrates vision input but has no v suffix,
  so it gets its own precise entry ahead of the text-only GLM patterns.

Without these entries the models fall through to the text-only default,
so createMediaContentPart() silently replaces attached images with a
text placeholder and isImageCapable() excludes the models from
vision-bridge candidates. Explicit modalities overrides were already
honored and are untouched.

Co-authored-by: Qwen-Coder <qwen-coder@alibabacloud.com>
---
 .../core/src/core/modalityDefaults.test.ts    | 34 +++++++++++++++++++
 packages/core/src/core/modalityDefaults.ts    | 11 ++++--
 2 files changed, 42 insertions(+), 3 deletions(-)

diff --git a/packages/core/src/core/modalityDefaults.test.ts b/packages/core/src/core/modalityDefaults.test.ts
index 8406363d901..82b00d3e5c9 100644
--- a/packages/core/src/core/modalityDefaults.test.ts
+++ b/packages/core/src/core/modalityDefaults.test.ts
@@ -202,6 +202,17 @@ describe('defaultModalities', () => {
     it('returns text-only for deepseek-reasoner', () => {
       expect(defaultModalities('deepseek-reasoner')).toEqual({});
     });
+
+    // (QwenLM/qwen-code#10270)
+    it('returns text-only for non-vision deepseek-v4-flash', () => {
+      expect(defaultModalities('deepseek-v4-flash')).toEqual({});
+    });
+
+    it('returns image for deepseek-v4-flash-vision-exp', () => {
+      const m = defaultModalities('deepseek-v4-flash-vision-exp');
+      expect(m.image).toBe(true);
+      expect(m.pdf).toBeUndefined();
+    });
   });
 
   describe('Zhipu GLM', () => {
@@ -211,6 +222,25 @@ describe('defaultModalities', () => {
       expect(m.pdf).toBeUndefined();
     });
 
+    // (QwenLM/qwen-code#10270)
+    it('returns image for glm-4.6v', () => {
+      const m = defaultModalities('glm-4.6v');
+      expect(m.image).toBe(true);
+      expect(m.pdf).toBeUndefined();
+    });
+
+    it('returns image for glm-5v-turbo', () => {
+      const m = defaultModalities('glm-5v-turbo');
+      expect(m.image).toBe(true);
+      expect(m.pdf).toBeUndefined();
+    });
+
+    it('returns image for glm-5.3-flash', () => {
+      const m = defaultModalities('glm-5.3-flash');
+      expect(m.image).toBe(true);
+      expect(m.pdf).toBeUndefined();
+    });
+
     it('returns text-only for glm-5', () => {
       expect(defaultModalities('glm-5')).toEqual({});
     });
@@ -218,6 +248,10 @@ describe('defaultModalities', () => {
     it('returns text-only for glm-4.7', () => {
       expect(defaultModalities('glm-4.7')).toEqual({});
     });
+
+    it('returns text-only for glm-4.6 (no v suffix)', () => {
+      expect(defaultModalities('glm-4.6')).toEqual({});
+    });
   });
 
   describe('MiniMax', () => {
diff --git a/packages/core/src/core/modalityDefaults.ts b/packages/core/src/core/modalityDefaults.ts
index d8adce67bb6..e70728688bc 100644
--- a/packages/core/src/core/modalityDefaults.ts
+++ b/packages/core/src/core/modalityDefaults.ts
@@ -61,14 +61,19 @@ const MODALITY_PATTERNS: Array<[RegExp, InputModalities]> = [
   [/^qwen/, {}],
 
   // -------------------
-  // DeepSeek — text-only
+  // DeepSeek — text-only, except explicit vision variants
+  // (QwenLM/qwen-code#10270)
   // -------------------
+  [/^deepseek-.*vision/, { image: true }],
   [/^deepseek/, {}],
 
   // -------------------
-  // Zhipu GLM
+  // Zhipu GLM — v-suffix ids are vision models; others are text-only
+  // (QwenLM/qwen-code#10270)
   // -------------------
-  [/^glm-4\.5v/, { image: true }],
+  [/^glm-[0-9.]+v/, { image: true }],
+  // glm-5.3-flash natively integrates vision input (no v suffix)
+  [/^glm-5\.3-flash/, { image: true }],
   [/^glm-5(?:-|$)/, {}],
   [/^glm-/, {}],