diff --git a/cli-config.yaml.example b/cli-config.yaml.example
index b1d7a67a9..0ccf75881 100644
--- a/cli-config.yaml.example
+++ b/cli-config.yaml.example
@@ -360,6 +360,7 @@ compression:
 #   # Defaults to Gemma 4 27B — natively multimodal, same model family as the main
 #   # text model, which avoids model-switching overhead and improves context continuity.
 #   # Override with any vision-capable model.  Set to "" to fall back to auto-detection.
+#   # Can also be overridden per-session with BROWSER_VISION_MODEL env var.
 #   browser_vision:
 #     model: "google/gemma-4-27b-it"  # default; override e.g. "google/gemini-2.5-flash"
 #
diff --git a/tests/tools/test_browser_vision_model.py b/tests/tools/test_browser_vision_model.py
index e565da89e..592de3a32 100644
--- a/tests/tools/test_browser_vision_model.py
+++ b/tests/tools/test_browser_vision_model.py
@@ -1,4 +1,12 @@
-"""Tests for browser_tool._get_vision_model() — Gemma 4 default (Issue #816)."""
+"""Tests for browser_tool._get_vision_model() — Gemma 4 default (Issue #816).
+
+Covers acceptance criteria from issue #816:
+- Browser screenshots use Gemma 4 by default.
+- BROWSER_VISION_MODEL env var overrides the model for browser vision only.
+- AUXILIARY_VISION_MODEL env var still works as a global override.
+- auxiliary.browser_vision.model in config.yaml overrides the default.
+- Priority: BROWSER_VISION_MODEL > config.yaml > AUXILIARY_VISION_MODEL > default.
+"""
 
 import importlib
 import sys
@@ -96,3 +104,26 @@ class TestGetVisionModelConfig:
         with patch("hermes_cli.config.load_config", return_value=cfg):
             import tools.browser_tool as bt
             assert bt._get_vision_model() == "openai/gpt-4o"
+
+    def test_config_beats_auxiliary_vision_model(self, monkeypatch):
+        """Config should override AUXILIARY_VISION_MODEL when BROWSER_VISION_MODEL unset."""
+        monkeypatch.delenv("BROWSER_VISION_MODEL", raising=False)
+        monkeypatch.setenv("AUXILIARY_VISION_MODEL", "global-override")
+        cfg = {"auxiliary": {"browser_vision": {"model": "config-model"}}}
+        with patch("hermes_cli.config.load_config", return_value=cfg):
+            import tools.browser_tool as bt
+            assert bt._get_vision_model() == "config-model"
+
+
+class TestBackwardCompatibility:
+    """AUXILIARY_VISION_MODEL must still work for users who already have it configured."""
+
+    def test_existing_auxiliary_vision_model_not_broken(self, monkeypatch):
+        """Users who set AUXILIARY_VISION_MODEL must not be broken by this change."""
+        monkeypatch.delenv("BROWSER_VISION_MODEL", raising=False)
+        monkeypatch.setenv("AUXILIARY_VISION_MODEL", "openai/gpt-4o")
+        import tools.browser_tool as bt
+        with patch("hermes_cli.config.load_config", return_value={}):
+            model = bt._get_vision_model()
+        assert model == "openai/gpt-4o"
+        assert model != "google/gemma-4-27b-it"
diff --git a/tools/browser_tool.py b/tools/browser_tool.py
index e2150c078..283549877 100644
--- a/tools/browser_tool.py
+++ b/tools/browser_tool.py
@@ -806,7 +806,7 @@ BROWSER_TOOL_SCHEMAS = [
     },
     {
         "name": "browser_vision",
-        "description": "Take a screenshot of the current page and analyze it with vision AI. Use this when you need to visually understand what's on the page - especially useful for CAPTCHAs, visual verification challenges, complex layouts, or when the text snapshot doesn't capture important visual information. Returns both the AI analysis and a screenshot_path that you can share with the user by including MEDIA:<screenshot_path> in your response. Requires browser_navigate to be called first.",
+        "description": "Take a screenshot of the current page and analyze it with vision AI (default: Gemma 4 multimodal). Use this when you need to visually understand what's on the page - especially useful for CAPTCHAs, visual verification challenges, complex layouts, or when the text snapshot doesn't capture important visual information. Returns both the AI analysis and a screenshot_path that you can share with the user by including MEDIA:<screenshot_path> in your response. Requires browser_navigate to be called first. Vision model can be overridden via BROWSER_VISION_MODEL env var or auxiliary.browser_vision.model in config.yaml.",
         "parameters": {
             "type": "object",
             "properties": {