feat(model-meta): register qwen-plus family with its own tokenizer and template

request_model values like qwen-plus matched no registered family, so
resolve_model_family fell through to its "glm5" default and the formatter
rendered canonical_prompt with the GLM-5 chat template and tokenized release
blocks with the GLM-5 tokenizer. The fallback is silent, so nothing surfaced it.

Register the family, ship Qwen3-plus tokenizer/chat_template (the template was
embedded in tokenizer_config.json upstream and is extracted here), and add a
QwenPlusTraceAdapter ahead of the GLM catch-all in get_raw_adapter -- the
registry entry alone is not enough, since the adapter is what picks the
template. The shared message/tool normalization is factored out of
build_glm5_canonical_prompt so both builders use one copy.

Measured against provider-reported usage.input_tokens on a 400-request sample,
median tokenized/reported ratio goes 0.9449 -> 1.0000, and unique release block
count on a 2190-request sample moves 5886 -> 5701.

Note deepseek-v4-flash is still unregistered and still resolves to glm5.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>

trace_formatter/raw_parser.py

@@ -57,6 +57,10 @@ def _load_qwen3_coder_chat_template():
     return _load_chat_template(str(resolve_chat_template_path("qwen3-coder")))
 
 
+def _load_qwen_plus_chat_template():
+    return _load_chat_template(str(resolve_chat_template_path("qwen-plus")))
+
+
 @lru_cache(maxsize=1)
 def _load_qwen3_coder_tool_parser_spec() -> Qwen3CoderToolParserSpec:
     defaults = Qwen3CoderToolParserSpec()
@@ -101,6 +105,10 @@ def _stringify_message_content_for_template(content):
 
 
 def _normalize_message_content_for_template(content, role=""):
+    if content is None:
+        return ""
+    if role == "tool" and not isinstance(content, (list, dict, str)):
+        return str(content)
     if role == "tool" and isinstance(content, (list, dict)):
         has_named_tool_refs = False
         if isinstance(content, list):
@@ -261,7 +269,35 @@ def _normalize_qwen_message_with_tool_parser(message, spec: Qwen3CoderToolParser
     return normalized_message
 
 
-def build_glm5_canonical_prompt(payload):
+_FALLBACK_COUNT = 0
+
+
+def _fallback_canonical_prompt(messages):
+    global _FALLBACK_COUNT
+    _FALLBACK_COUNT += 1
+    parts = []
+    for m in messages:
+        if not isinstance(m, dict):
+            parts.append(str(m))
+            continue
+        role = str(m.get("role", ""))
+        content = _stringify_message_content_for_template(m.get("content"))
+        seg = "<|" + role + "|>" + content
+        tool_calls = m.get("tool_calls") or []
+        if isinstance(tool_calls, list):
+            for tc in tool_calls:
+                function = tc.get("function", {}) if isinstance(tc, dict) else {}
+                if not isinstance(function, dict):
+                    function = {}
+                name = str(function.get("name", ""))
+                arguments = function.get("arguments", {})
+                seg += "<tool_call>" + name + compact_json(arguments) + "</tool_call>"
+        parts.append(seg)
+    return "".join(parts)
+
+
+def _normalize_payload_for_template(payload):
+    """Normalize one request payload into (messages, tools) ready for a chat template."""
     input_payload = payload.get("input", {}) if isinstance(payload, dict) else {}
     parameters = payload.get("parameters", {}) if isinstance(payload, dict) else {}
     messages = []
@@ -289,11 +325,33 @@ def build_glm5_canonical_prompt(payload):
     elif not isinstance(tools_payload, list):
         tools_payload = [tools_payload]
     tools = [_normalize_tool_spec_for_template(tool) for tool in tools_payload if isinstance(tool, dict)]
-    return _load_glm5_chat_template().render(
-        messages=messages,
-        tools=tools,
-        add_generation_prompt=True,
-    )
+    return messages, tools
+
+
+def build_glm5_canonical_prompt(payload):
+    messages, tools = _normalize_payload_for_template(payload)
+    try:
+        return _load_glm5_chat_template().render(
+            messages=messages,
+            tools=tools,
+            add_generation_prompt=True,
+        )
+    except Exception:
+        return _fallback_canonical_prompt(messages)
+
+
+def build_qwen_plus_canonical_prompt(payload):
+    messages, tools = _normalize_payload_for_template(payload)
+    if not messages:
+        messages = [{"role": "system", "content": ""}]
+    try:
+        return _load_qwen_plus_chat_template().render(
+            messages=messages,
+            tools=tools,
+            add_generation_prompt=True,
+        )
+    except Exception:
+        return _fallback_canonical_prompt(messages)
 
 
 def build_qwen3_coder_canonical_prompt(payload):
@@ -408,8 +466,20 @@ class QwenTraceAdapter(GLMTraceAdapter):
         return build_qwen3_coder_canonical_prompt(payload)
 
 
+class QwenPlusTraceAdapter(GLMTraceAdapter):
+    name = "qwen-plus"
+
+    def detect(self, raw):
+        if "request_params" not in raw:
+            return False
+        return infer_model_family_from_request_model(raw.get("request_model")) == "qwen-plus"
+
+    def build_canonical_prompt(self, payload):
+        return build_qwen_plus_canonical_prompt(payload)
+
+
 def get_raw_adapter(raw: dict[str, Any]):
-    for adapter in [QwenTraceAdapter(), GLMTraceAdapter()]:
+    for adapter in [QwenTraceAdapter(), QwenPlusTraceAdapter(), GLMTraceAdapter()]:
         if adapter.detect(raw):
             return adapter
     raise ValueError("Unsupported raw trace format for trace_formatter.")