feat: Track total tokens and use in benchmark stats

aider/coders/base_coder.py

@@ -108,8 +108,6 @@ class Coder:
     partial_response_content = ""
     commit_before_message = []
     message_cost = 0.0
-    message_tokens_sent = 0
-    message_tokens_received = 0
     add_cache_headers = False
     cache_warming_thread = None
     num_cache_warming_pings = 0
@@ -128,6 +126,8 @@ class Coder:
         from_coder=None,
         summarize_from_coder=True,
         **kwargs,
+
+
     ):
         import aider.coders as coders
 
@@ -175,6 +175,8 @@ class Coder:
                 commands=from_coder.commands.clone(),
                 total_cost=from_coder.total_cost,
                 ignore_mentions=from_coder.ignore_mentions,
+                total_tokens_sent=from_coder.total_tokens_sent,
+                total_tokens_received=from_coder.total_tokens_received,
                 file_watcher=from_coder.file_watcher,
             )
             use_kwargs.update(update)  # override to complete the switch
@@ -327,6 +329,8 @@ class Coder:
         chat_language=None,
         detect_urls=True,
         ignore_mentions=None,
+        total_tokens_sent=0,
+        total_tokens_received=0,
         file_watcher=None,
         auto_copy_context=False,
         auto_accept_architect=True,
@@ -373,6 +377,10 @@ class Coder:
         self.need_commit_before_edits = set()
 
         self.total_cost = total_cost
+        self.total_tokens_sent = total_tokens_sent
+        self.total_tokens_received = total_tokens_received
+        self.message_tokens_sent = 0
+        self.message_tokens_received = 0
 
         self.verbose = verbose
         self.abs_fnames = set()
@@ -2057,6 +2065,9 @@ class Coder:
         if not self.usage_report:
             return
 
+        self.total_tokens_sent += self.message_tokens_sent
+        self.total_tokens_received += self.message_tokens_received
+
         self.io.tool_output(self.usage_report)
 
         prompt_tokens = self.message_tokens_sent

benchmark/benchmark.py

@@ -957,8 +957,8 @@ def run_test_real(
         indentation_errors=indentation_errors,
         lazy_comments=lazy_comments,  # Add the count of pattern matches to the results
         reasoning_effort=reasoning_effort,
-        prompt_tokens=coder.message_tokens_sent,
-        completion_tokens=coder.message_tokens_received,
+        prompt_tokens=coder.total_tokens_sent,
+        completion_tokens=coder.total_tokens_received,
         thinking_tokens=thinking_tokens,
         chat_hashes=list(
             zip(