diff --git a/ollamallmbasev3/Chart.yaml b/ollamallmbasev3/Chart.yaml
index 9cfafc95f..5450c1b82 100644
--- a/ollamallmbasev3/Chart.yaml
+++ b/ollamallmbasev3/Chart.yaml
@@ -1,6 +1,6 @@
 apiVersion: v2
-appVersion: 0.32.1
+appVersion: 0.32.9
 description: Generic Ollama + llm-init base; the model is supplied at install time via env
 name: ollamallmbasev3
 type: application
-version: 1.3.10
+version: 1.3.11
diff --git a/ollamallmbasev3/OlaresManifest.yaml b/ollamallmbasev3/OlaresManifest.yaml
index f06d1e9f2..394f7cb1f 100644
--- a/ollamallmbasev3/OlaresManifest.yaml
+++ b/ollamallmbasev3/OlaresManifest.yaml
@@ -7,7 +7,7 @@ metadata:
   description: "Generic Ollama engine base. Pick any Ollama library model at install via env."
   appid: ollamallmbasev3
   title: Ollama Engine Base
-  version: '1.3.10'
+  version: '1.3.11'
   categories:
     - AI
 sharedEntrances:
@@ -33,19 +33,30 @@ workloadReplicas:
   llminit: 1
 spec:
   onlyAdmin: true
-  versionName: '0.32.1'
+  versionName: '0.32.9'
   upgradeDescription: |
-    Upgrade Ollama from v0.31.2 to v0.32.1. Chart 1.3.10.
+    Upgrade Ollama from v0.32.1 to v0.32.9. Chart 1.3.11.
 
     **What's Changed** (relevant to Olares)
 
-    v0.32.1
-    - Improved Gemma 4 tool calling and multi-turn reasoning, including more reliable tool-response continuations
+    v0.32.9
+    - Added Nemotron 3 architecture (NVIDIA Nemotron 3.5 Lightning MoE support)
+    - Fixed Muse Glimmer function-calling parser boundary condition
 
-    v0.32.0
-    - No server/runtime changes affecting Olares (release focused on interactive agent CLI and `ollama launch` integrations)
+    v0.32.8
+    - Muse Glimmer now supported on NVIDIA, AMD, and additional platforms
 
-    Drop accelerator mode `cpu` (GPU-only: `nvidia` / `nvidia-gb10`). Chart 1.3.9.
+    v0.32.6
+    - `/v1/chat/completions` streaming matches OpenAI wire format (`role` on first chunk only, `finish_reason` and usage in separate chunks)
+    - Truncated responses report `finish_reason: "length"` instead of `"tool_calls"`
+
+    v0.32.4
+    - Fixed Qwen3 MoE decoding for differently-quantized experts
+
+    v0.32.3
+    - Fixed model downloads that stall before the first byte
+    - Fixed GLM tool calls silently dropped at end of generation
+    - Updated llama.cpp and MLX engines
 
     For detailed release notes, visit:
     https://github.com/ollama/ollama/releases
diff --git a/ollamallmbasev3/i18n/en-US/OlaresManifest.yaml b/ollamallmbasev3/i18n/en-US/OlaresManifest.yaml
index c08177030..5fb1c6062 100644
--- a/ollamallmbasev3/i18n/en-US/OlaresManifest.yaml
+++ b/ollamallmbasev3/i18n/en-US/OlaresManifest.yaml
@@ -4,17 +4,28 @@ metadata:
 
 spec:
   upgradeDescription: |
-    Upgrade Ollama from v0.31.2 to v0.32.1. Chart 1.3.10.
+    Upgrade Ollama from v0.32.1 to v0.32.9. Chart 1.3.11.
 
     **What's Changed** (relevant to Olares)
 
-    v0.32.1
-    - Improved Gemma 4 tool calling and multi-turn reasoning, including more reliable tool-response continuations
+    v0.32.9
+    - Added Nemotron 3 architecture (NVIDIA Nemotron 3.5 Lightning MoE support)
+    - Fixed Muse Glimmer function-calling parser boundary condition
 
-    v0.32.0
-    - No server/runtime changes affecting Olares (release focused on interactive agent CLI and `ollama launch` integrations)
+    v0.32.8
+    - Muse Glimmer now supported on NVIDIA, AMD, and additional platforms
 
-    Drop accelerator mode `cpu` (GPU-only: `nvidia` / `nvidia-gb10`). Chart 1.3.9.
+    v0.32.6
+    - `/v1/chat/completions` streaming matches OpenAI wire format (`role` on first chunk only, `finish_reason` and usage in separate chunks)
+    - Truncated responses report `finish_reason: "length"` instead of `"tool_calls"`
+
+    v0.32.4
+    - Fixed Qwen3 MoE decoding for differently-quantized experts
+
+    v0.32.3
+    - Fixed model downloads that stall before the first byte
+    - Fixed GLM tool calls silently dropped at end of generation
+    - Updated llama.cpp and MLX engines
 
     For detailed release notes, visit:
     https://github.com/ollama/ollama/releases
diff --git a/ollamallmbasev3/i18n/zh-CN/OlaresManifest.yaml b/ollamallmbasev3/i18n/zh-CN/OlaresManifest.yaml
index 23eb36828..c050cf405 100644
--- a/ollamallmbasev3/i18n/zh-CN/OlaresManifest.yaml
+++ b/ollamallmbasev3/i18n/zh-CN/OlaresManifest.yaml
@@ -4,17 +4,28 @@ metadata:
 
 spec:
   upgradeDescription: |
-    将 Ollama 从 v0.31.2 升级至 v0.32.1。Chart 1.3.10。
+    将 Ollama 从 v0.32.1 升级至 v0.32.9。Chart 1.3.11。
 
     **更新内容**（与 Olares 相关）
 
-    v0.32.1
-    - 改进 Gemma 4 工具调用与多轮推理，工具响应续写更可靠
+    v0.32.9
+    - 新增 Nemotron 3 架构（支持 NVIDIA Nemotron 3.5 Lightning MoE）
+    - 修复 Muse Glimmer 工具调用解析器的边界条件
 
-    v0.32.0
-    - 无影响 Olares 的服务端/运行时变更（本版本主要面向 interactive agent CLI 与 `ollama launch` 集成）
+    v0.32.8
+    - Muse Glimmer 现已支持 NVIDIA、AMD 及其他平台
 
-    移除加速器模式 `cpu`（仅保留 GPU：`nvidia` / `nvidia-gb10`）。Chart 1.3.9。
+    v0.32.6
+    - `/v1/chat/completions` 流式输出对齐 OpenAI 格式（`role` 仅首块、`finish_reason` 与 usage 独立分块）
+    - 截断响应的 `finish_reason` 由 `"tool_calls"` 修正为 `"length"`
+
+    v0.32.4
+    - 修复 Qwen3 MoE 不同量化 expert 的解码问题
+
+    v0.32.3
+    - 修复模型下载在首字节前卡住的问题
+    - 修复 GLM 工具调用在生成末尾被静默丢弃
+    - 更新 llama.cpp 与 MLX 引擎
 
     详细发布说明请访问：
     https://github.com/ollama/ollama/releases
diff --git a/ollamallmbasev3/templates/ollama.yaml b/ollamallmbasev3/templates/ollama.yaml
index 13ca8a70f..7a6c955fe 100644
--- a/ollamallmbasev3/templates/ollama.yaml
+++ b/ollamallmbasev3/templates/ollama.yaml
@@ -84,7 +84,7 @@ spec:
             defaultMode: 0555
       containers:
         - name: ollama
-          image: docker.io/ollama/ollama:0.32.1
+          image: docker.io/ollama/ollama:0.32.9
           imagePullPolicy: IfNotPresent
           command: ["/bin/sh", "/llm-init/wrappers/ollama.sh"]
           env: