修复 Completions转Anthropic时不记录实际返回模型、Input token记录错误问题 (#2774)

* fix(proxy): 修复completions转claude格式流式响应未记录实际命中模型

* style: cargo fmt fix

* fix(proxy): 修复completions转claude格式时input与cache_read重复计费

* fix(proxy): 修复完全缓存命中时input_tokens计算错误

* test: 更新input_tokens期望值匹配去重逻辑

Author: LaoYueHanNi

src-tauri/src/proxy/handlers.rs

@@ -333,7 +333,7 @@ async fn handle_claude_transform(
         let usage_collector = if usage_logging_enabled(state) {
             let state = state.clone();
             let provider_id = ctx.provider.id.clone();
-            let model = ctx.request_model.clone();
+            let request_model = ctx.request_model.clone();
             let status_code = status.as_u16();
             let start_time = ctx.start_time;
             let session_id = ctx.session_id.clone();
@@ -343,19 +343,20 @@ async fn handle_claude_transform(
                 Some(claude_stream_usage_event_filter),
                 move |events, first_token_ms| {
                     if let Some(usage) = TokenUsage::from_claude_stream_events(&events) {
+                        let model = usage.model.clone().unwrap_or(request_model.clone());
                         let latency_ms = start_time.elapsed().as_millis() as u64;
                         let state = state.clone();
                         let provider_id = provider_id.clone();
-                        let model = model.clone();
                         let session_id = session_id.clone();
+                        let request_model = request_model.clone();
 
                         tokio::spawn(async move {
                             log_usage(
                                 &state,
                                 &provider_id,
                                 "claude",
                                 &model,
-                                &model,
+                                &request_model,
                                 usage,
                                 latency_ms,
                                 first_token_ms,

src-tauri/src/proxy/providers/streaming.rs

@@ -100,11 +100,14 @@ struct ToolBlockState {
 const INFINITE_WHITESPACE_THRESHOLD: usize = 500;
 
 fn build_anthropic_usage_json(usage: &Usage) -> Value {
+    // OpenAI prompt_tokens 含缓存，Anthropic input_tokens 不含，需减去
+    let cached = extract_cache_read_tokens(usage).unwrap_or(0);
+    let input_tokens = usage.prompt_tokens.saturating_sub(cached);
     let mut usage_json = json!({
-        "input_tokens": usage.prompt_tokens,
+        "input_tokens": input_tokens,
         "output_tokens": usage.completion_tokens
     });
-    if let Some(cached) = extract_cache_read_tokens(usage) {
+    if cached > 0 {
         usage_json["cache_read_input_tokens"] = json!(cached);
     }
     if let Some(created) = usage.cache_creation_input_tokens {
@@ -223,8 +226,10 @@ pub fn create_anthropic_sse_stream<E: std::error::Error + Send + 'static>(
                                                 "output_tokens": 0
                                             });
                                             if let Some(u) = &chunk.usage {
-                                                start_usage["input_tokens"] = json!(u.prompt_tokens);
-                                                if let Some(cached) = extract_cache_read_tokens(u) {
+                                                let cached = extract_cache_read_tokens(u).unwrap_or(0);
+                                                let input = u.prompt_tokens.saturating_sub(cached);
+                                                start_usage["input_tokens"] = json!(input);
+                                                if cached > 0 {
                                                     start_usage["cache_read_input_tokens"] = json!(cached);
                                                 }
                                                 if let Some(created) = u.cache_creation_input_tokens {
@@ -1022,7 +1027,7 @@ mod tests {
             message_delta
                 .pointer("/usage/input_tokens")
                 .and_then(|v| v.as_u64()),
-            Some(13312)
+            Some(13212)
         );
         assert_eq!(
             message_delta