feat: implement reasoning_tokens tracking and enhanced usage logging

src/providers/gemini.rs

@@ -722,6 +722,10 @@ impl super::Provider for GeminiProvider {
         let reasoning_content = candidate
             .and_then(|c| c.content.parts.iter().find_map(|p| p.thought.clone()));
 
+        let reasoning_tokens = reasoning_content.as_ref()
+            .map(|r| crate::utils::tokens::estimate_completion_tokens(r, &model))
+            .unwrap_or(0);
+
         // Extract function calls → OpenAI tool_calls
         let tool_calls = candidate.and_then(|c| Self::extract_tool_calls(&c.content.parts));
 
@@ -752,6 +756,7 @@ impl super::Provider for GeminiProvider {
             tool_calls,
             prompt_tokens,
             completion_tokens,
+            reasoning_tokens,
             total_tokens,
             cache_read_tokens,
             cache_write_tokens: 0, // Gemini doesn't report cache writes separately
@@ -902,6 +907,7 @@ impl super::Provider for GeminiProvider {
                             super::StreamUsage {
                                 prompt_tokens: u.prompt_token_count,
                                 completion_tokens: u.candidates_token_count,
+                                reasoning_tokens: 0,
                                 total_tokens: u.total_token_count,
                                 cache_read_tokens: u.cached_content_token_count,
                                 cache_write_tokens: 0,

src/providers/helpers.rs

@@ -254,6 +254,11 @@ pub fn parse_openai_response(resp_json: &Value, model: String) -> Result<Provide
     let completion_tokens = usage["completion_tokens"].as_u64().unwrap_or(0) as u32;
     let total_tokens = usage["total_tokens"].as_u64().unwrap_or(0) as u32;
 
+    // Extract reasoning tokens
+    let reasoning_tokens = usage["completion_tokens_details"]["reasoning_tokens"]
+        .as_u64()
+        .unwrap_or(0) as u32;
+
     // Extract cache tokens — try OpenAI/Grok format first, then DeepSeek format
     let cache_read_tokens = usage["prompt_tokens_details"]["cached_tokens"]
         .as_u64()
@@ -271,6 +276,7 @@ pub fn parse_openai_response(resp_json: &Value, model: String) -> Result<Provide
         tool_calls,
         prompt_tokens,
         completion_tokens,
+        reasoning_tokens,
         total_tokens,
         cache_read_tokens,
         cache_write_tokens,
@@ -295,6 +301,10 @@ pub fn parse_openai_stream_chunk(
         let completion_tokens = u["completion_tokens"].as_u64().unwrap_or(0) as u32;
         let total_tokens = u["total_tokens"].as_u64().unwrap_or(0) as u32;
 
+        let reasoning_tokens = u["completion_tokens_details"]["reasoning_tokens"]
+            .as_u64()
+            .unwrap_or(0) as u32;
+
         let cache_read_tokens = u["prompt_tokens_details"]["cached_tokens"]
             .as_u64()
             .or_else(|| u["prompt_cache_hit_tokens"].as_u64())
@@ -305,6 +315,7 @@ pub fn parse_openai_stream_chunk(
         Some(StreamUsage {
             prompt_tokens,
             completion_tokens,
+            reasoning_tokens,
             total_tokens,
             cache_read_tokens,
             cache_write_tokens,

src/providers/mod.rs

@@ -75,6 +75,7 @@ pub struct ProviderResponse {
     pub tool_calls: Option<Vec<crate::models::ToolCall>>,
     pub prompt_tokens: u32,
     pub completion_tokens: u32,
+    pub reasoning_tokens: u32,
     pub total_tokens: u32,
     pub cache_read_tokens: u32,
     pub cache_write_tokens: u32,
@@ -86,6 +87,7 @@ pub struct ProviderResponse {
 pub struct StreamUsage {
     pub prompt_tokens: u32,
     pub completion_tokens: u32,
+    pub reasoning_tokens: u32,
     pub total_tokens: u32,
     pub cache_read_tokens: u32,
     pub cache_write_tokens: u32,

src/providers/openai.rs

@@ -177,6 +177,7 @@ impl super::Provider for OpenAIProvider {
                     tool_calls: None,
                     prompt_tokens,
                     completion_tokens,
+                    reasoning_tokens: 0,
                     total_tokens,
                     cache_read_tokens: 0,
                     cache_write_tokens: 0,
@@ -275,6 +276,7 @@ impl super::Provider for OpenAIProvider {
             tool_calls: None,
             prompt_tokens,
             completion_tokens,
+            reasoning_tokens: 0,
             total_tokens,
             cache_read_tokens: 0,
             cache_write_tokens: 0,