feat(usage): 在 CanonicalUsage 中追加缓存 token 字段并在标准化流程透传

- 扩展 CanonicalUsage 支持 cache_creation/cache_read 及 5m/1h 明细
- OpenAI/Claude/Gemini 解析器提取缓存 token 并计入 total
- usage_mapper 补齐 Claude 缓存字段映射
- 新增 gateway pricing 集成测试覆盖三家同步/流式缓存计费

crates/aether-ai-pipeline/src/finalize/standard/gemini/stream.rs

@@ -514,6 +514,7 @@ mod tests {
                             input_tokens: 1,
                             output_tokens: 2,
                             total_tokens: 3,
+                            ..CanonicalUsage::default()
                         }),
                     },
                 })

crates/aether-ai-pipeline/src/finalize/standard/openai/stream.rs

@@ -1764,6 +1764,7 @@ mod tests {
                     input_tokens: 1,
                     output_tokens: 2,
                     total_tokens: 3,
+                    ..CanonicalUsage::default()
                 }),
             },
         };
@@ -1902,6 +1903,7 @@ mod tests {
                     input_tokens: 1,
                     output_tokens: 2,
                     total_tokens: 3,
+                    ..
                 }),
             } if reason == "tool_calls"
         )));
@@ -1961,6 +1963,7 @@ mod tests {
                             input_tokens: 1,
                             output_tokens: 2,
                             total_tokens: 3,
+                            ..CanonicalUsage::default()
                         }),
                     },
                 })
@@ -2043,6 +2046,7 @@ mod tests {
                     input_tokens: 1,
                     output_tokens: 2,
                     total_tokens: 3,
+                    ..
                 }),
             } if reason == "stop"
         )));
@@ -2078,6 +2082,7 @@ mod tests {
                             input_tokens: 1,
                             output_tokens: 2,
                             total_tokens: 3,
+                            ..CanonicalUsage::default()
                         }),
                     },
                 })

crates/aether-ai-pipeline/src/finalize/standard/stream_core/common.rs

@@ -5,6 +5,10 @@ pub struct CanonicalUsage {
     pub input_tokens: u64,
     pub output_tokens: u64,
     pub total_tokens: u64,
+    pub cache_creation_tokens: u64,
+    pub cache_creation_ephemeral_5m_tokens: u64,
+    pub cache_creation_ephemeral_1h_tokens: u64,
+    pub cache_read_tokens: u64,
 }
 
 #[derive(Clone, Debug)]
@@ -78,14 +82,43 @@ pub fn canonical_usage_from_openai_usage(value: Option<&Value>) -> Option<Canoni
         .or_else(|| usage.get("completion_tokens"))
         .and_then(Value::as_u64)
         .unwrap_or(0);
-    let total_tokens = usage
-        .get("total_tokens")
+    let cache_creation_tokens = usage
+        .get("cache_creation_input_tokens")
         .and_then(Value::as_u64)
-        .unwrap_or(input_tokens + output_tokens);
+        .or_else(|| {
+            usage
+                .get("input_tokens_details")
+                .or_else(|| usage.get("prompt_tokens_details"))
+                .and_then(Value::as_object)
+                .and_then(|details| details.get("cached_creation_tokens"))
+                .and_then(Value::as_u64)
+        })
+        .unwrap_or(0);
+    let cache_read_tokens = usage
+        .get("cache_read_input_tokens")
+        .and_then(Value::as_u64)
+        .or_else(|| {
+            usage
+                .get("input_tokens_details")
+                .or_else(|| usage.get("prompt_tokens_details"))
+                .and_then(Value::as_object)
+                .and_then(|details| details.get("cached_tokens"))
+                .and_then(Value::as_u64)
+        })
+        .unwrap_or(0);
+    let total_tokens = usage.get("total_tokens").and_then(Value::as_u64).unwrap_or(
+        input_tokens
+            .saturating_add(output_tokens)
+            .saturating_add(cache_creation_tokens)
+            .saturating_add(cache_read_tokens),
+    );
     Some(CanonicalUsage {
         input_tokens,
         output_tokens,
         total_tokens,
+        cache_creation_tokens,
+        cache_read_tokens,
+        ..CanonicalUsage::default()
     })
 }
 
@@ -99,10 +132,39 @@ pub fn canonical_usage_from_claude_usage(value: Option<&Value>) -> Option<Canoni
         .get("output_tokens")
         .and_then(Value::as_u64)
         .unwrap_or(0);
+    let cache_creation_ephemeral_5m_tokens = usage
+        .get("cache_creation")
+        .and_then(Value::as_object)
+        .and_then(|value| value.get("ephemeral_5m_input_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let cache_creation_ephemeral_1h_tokens = usage
+        .get("cache_creation")
+        .and_then(Value::as_object)
+        .and_then(|value| value.get("ephemeral_1h_input_tokens"))
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
+    let cache_creation_tokens = usage
+        .get("cache_creation_input_tokens")
+        .and_then(Value::as_u64)
+        .unwrap_or(
+            cache_creation_ephemeral_5m_tokens.saturating_add(cache_creation_ephemeral_1h_tokens),
+        );
+    let cache_read_tokens = usage
+        .get("cache_read_input_tokens")
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
     Some(CanonicalUsage {
         input_tokens,
         output_tokens,
-        total_tokens: input_tokens + output_tokens,
+        total_tokens: input_tokens
+            .saturating_add(output_tokens)
+            .saturating_add(cache_creation_tokens)
+            .saturating_add(cache_read_tokens),
+        cache_creation_tokens,
+        cache_creation_ephemeral_5m_tokens,
+        cache_creation_ephemeral_1h_tokens,
+        cache_read_tokens,
     })
 }
 
@@ -116,14 +178,24 @@ pub fn canonical_usage_from_gemini_usage(value: Option<&Value>) -> Option<Canoni
         .get("candidatesTokenCount")
         .and_then(Value::as_u64)
         .unwrap_or(0);
+    let cache_read_tokens = usage
+        .get("cachedContentTokenCount")
+        .and_then(Value::as_u64)
+        .unwrap_or(0);
     let total_tokens = usage
         .get("totalTokenCount")
         .and_then(Value::as_u64)
-        .unwrap_or(input_tokens + output_tokens);
+        .unwrap_or(
+            input_tokens
+                .saturating_add(output_tokens)
+                .saturating_add(cache_read_tokens),
+        );
     Some(CanonicalUsage {
         input_tokens,
         output_tokens,
         total_tokens,
+        cache_read_tokens,
+        ..CanonicalUsage::default()
     })
 }
 

crates/aether-ai-pipeline/src/finalize/standard/stream_core/format_matrix.rs

@@ -259,11 +259,17 @@ fn standardized_usage_from_canonical(usage: CanonicalUsage) -> StandardizedUsage
     let mut standardized = StandardizedUsage::new();
     standardized.input_tokens = usage.input_tokens as i64;
     standardized.output_tokens = usage.output_tokens as i64;
+    standardized.cache_creation_tokens = usage.cache_creation_tokens as i64;
+    standardized.cache_creation_ephemeral_5m_tokens =
+        usage.cache_creation_ephemeral_5m_tokens as i64;
+    standardized.cache_creation_ephemeral_1h_tokens =
+        usage.cache_creation_ephemeral_1h_tokens as i64;
+    standardized.cache_read_tokens = usage.cache_read_tokens as i64;
     standardized.dimensions.insert(
         "total_tokens".to_string(),
         serde_json::json!(usage.total_tokens),
     );
-    standardized
+    standardized.normalize_cache_creation_breakdown()
 }
 
 impl ClientStreamEmitter {
@@ -413,7 +419,7 @@ fn parse_gemini_error(payload: &Value) -> Option<(String, Option<String>, LocalC
 
 #[cfg(test)]
 mod tests {
-    use super::StreamingStandardFormatMatrix;
+    use super::{StreamingStandardFormatMatrix, StreamingStandardTerminalObserver};
     use serde_json::{json, Value};
 
     fn report_context(provider_api_format: &str, client_api_format: &str) -> Value {
@@ -689,4 +695,53 @@ mod tests {
                 .is_empty());
         }
     }
+
+    #[test]
+    fn terminal_observer_preserves_claude_cache_usage() {
+        let report_context = report_context("claude:chat", "openai:chat");
+        let mut observer = StreamingStandardTerminalObserver::default();
+
+        observer
+            .push_line(
+                &report_context,
+                data_line(json!({
+                    "type": "message_start",
+                    "message": {
+                        "id": "msg_cache_123",
+                        "model": "claude-sonnet-4-5"
+                    }
+                })),
+            )
+            .expect("message_start should parse");
+        observer
+            .push_line(
+                &report_context,
+                data_line(json!({
+                    "type": "message_delta",
+                    "delta": {
+                        "stop_reason": "end_turn"
+                    },
+                    "usage": {
+                        "input_tokens": 6,
+                        "output_tokens": 20,
+                        "cache_creation_input_tokens": 42262,
+                        "cache_read_input_tokens": 0
+                    }
+                })),
+            )
+            .expect("message_delta should parse");
+
+        let summary = observer
+            .latest_summary()
+            .cloned()
+            .expect("summary should exist");
+        let usage = summary
+            .standardized_usage
+            .expect("standardized usage should exist");
+
+        assert_eq!(usage.input_tokens, 6);
+        assert_eq!(usage.output_tokens, 20);
+        assert_eq!(usage.cache_creation_tokens, 42_262);
+        assert_eq!(usage.cache_read_tokens, 0);
+    }
 }

crates/aether-usage-runtime/src/usage_mapper.rs

@@ -66,6 +66,22 @@ fn base_mapping(api_format: &str) -> BTreeMap<String, String> {
             mapping.insert("completion_tokens".to_string(), "output_tokens".to_string());
             mapping.insert("input_tokens".to_string(), "input_tokens".to_string());
             mapping.insert("output_tokens".to_string(), "output_tokens".to_string());
+            mapping.insert(
+                "cache_creation_input_tokens".to_string(),
+                "cache_creation_tokens".to_string(),
+            );
+            mapping.insert(
+                "cache_creation.ephemeral_5m_input_tokens".to_string(),
+                "cache_creation_ephemeral_5m_tokens".to_string(),
+            );
+            mapping.insert(
+                "cache_creation.ephemeral_1h_input_tokens".to_string(),
+                "cache_creation_ephemeral_1h_tokens".to_string(),
+            );
+            mapping.insert(
+                "cache_read_input_tokens".to_string(),
+                "cache_read_tokens".to_string(),
+            );
             mapping.insert(
                 "prompt_tokens_details.cached_tokens".to_string(),
                 "cache_read_tokens".to_string(),
@@ -252,6 +268,26 @@ mod tests {
         assert_eq!(usage.reasoning_tokens, 1);
     }
 
+    #[test]
+    fn maps_openai_responses_with_top_level_cache_fields() {
+        let usage = map_usage_from_response(
+            &serde_json::json!({
+                "usage": {
+                    "input_tokens": 6,
+                    "output_tokens": 20,
+                    "cache_creation_input_tokens": 42_262,
+                    "cache_read_input_tokens": 0
+                }
+            }),
+            "openai:chat",
+        );
+
+        assert_eq!(usage.input_tokens, 6);
+        assert_eq!(usage.output_tokens, 20);
+        assert_eq!(usage.cache_creation_tokens, 42_262);
+        assert_eq!(usage.cache_read_tokens, 0);
+    }
+
     #[test]
     fn maps_openai_responses_usage_from_stream_chunks() {
         let usage = map_usage_from_response(