fix(qwen): correct context lengths for qwen3-coder models and send max_tokens to portal

Based on PR #7285 by @kshitijk4poor.

Two bugs affecting Qwen OAuth users:

1. Wrong context window — qwen3-coder-plus showed 128K instead of 1M.
   Added specific entries before the generic qwen catch-all:
   - qwen3-coder-plus: 1,000,000 (corrected from PR's 1,048,576 per
     official Alibaba Cloud docs and OpenRouter)
   - qwen3-coder: 262,144

2. Random stopping — max_tokens was suppressed for Qwen Portal, so the
   server applied its own low default. Reasoning models exhaust that on
   thinking tokens. Now: honor explicit max_tokens, default to 65536
   when unset.

Co-authored-by: kshitijk4poor <82637225+kshitijk4poor@users.noreply.github.com>

tests/agent/test_model_metadata.py

@@ -222,6 +222,24 @@ class TestGetModelContextLength:
         mock_fetch.return_value = {}
         assert get_model_context_length("openai/gpt-4o") == 128000
 
+    @patch("agent.model_metadata.fetch_model_metadata")
+    def test_qwen3_coder_plus_context_length(self, mock_fetch):
+        """qwen3-coder-plus has a 1M context window, not the generic 128K Qwen default."""
+        mock_fetch.return_value = {}
+        assert get_model_context_length("qwen3-coder-plus") == 1000000
+
+    @patch("agent.model_metadata.fetch_model_metadata")
+    def test_qwen3_coder_context_length(self, mock_fetch):
+        """qwen3-coder has a 256K context window, not the generic 128K Qwen default."""
+        mock_fetch.return_value = {}
+        assert get_model_context_length("qwen3-coder") == 262144
+
+    @patch("agent.model_metadata.fetch_model_metadata")
+    def test_qwen_generic_context_length(self, mock_fetch):
+        """Generic qwen models still get the 128K default."""
+        mock_fetch.return_value = {}
+        assert get_model_context_length("qwen3-plus") == 131072
+
     @patch("agent.model_metadata.fetch_model_metadata")
     def test_api_missing_context_length_key(self, mock_fetch):
         """Model in API but without context_length → defaults to 128000."""

tests/run_agent/test_run_agent.py

@@ -953,14 +953,24 @@ class TestBuildApiKwargs:
         assert kwargs["messages"][0]["content"][0]["text"] == "hi"
         assert "cache_control" not in kwargs["messages"][0]["content"][0]
 
-    def test_qwen_portal_omits_max_tokens(self, agent):
+    def test_qwen_portal_sends_explicit_max_tokens(self, agent):
+        """When the user explicitly sets max_tokens, it should be sent to Qwen Portal."""
         agent.base_url = "https://portal.qwen.ai/v1"
         agent._base_url_lower = agent.base_url.lower()
         agent.max_tokens = 4096
         messages = [{"role": "system", "content": "sys"}, {"role": "user", "content": "hi"}]
         kwargs = agent._build_api_kwargs(messages)
-        assert "max_tokens" not in kwargs
-        assert "max_completion_tokens" not in kwargs
+        assert kwargs["max_tokens"] == 4096
+
+    def test_qwen_portal_default_max_tokens(self, agent):
+        """When max_tokens is None, Qwen Portal gets a default of 65536
+        to prevent reasoning models from exhausting their output budget."""
+        agent.base_url = "https://portal.qwen.ai/v1"
+        agent._base_url_lower = agent.base_url.lower()
+        agent.max_tokens = None
+        messages = [{"role": "system", "content": "sys"}, {"role": "user", "content": "hi"}]
+        kwargs = agent._build_api_kwargs(messages)
+        assert kwargs["max_tokens"] == 65536
 
 
 class TestBuildAssistantMessage: