관리자 사용량 근거를 정교화

apps/api/app/services/llm_pricing.py

@@ -9,10 +9,14 @@
 from __future__ import annotations
 
 from dataclasses import dataclass
+from datetime import date, datetime, timezone
 
 
 MILLION = 1_000_000
 RATE_CARD_VERSION = "2026-07-31"
+GOOGLE_36_RELEASE_START = date(2026, 7, 21)
+GOOGLE_FLASH_INTRO_START = date(2026, 8, 13)
+GOOGLE_FLASH_INTRO_END = date(2027, 1, 1)
 
 GOOGLE_PRICING_URL = "https://ai.google.dev/gemini-api/docs/pricing"
 OPENAI_CODEX_RATE_URL = "https://help.openai.com/en/articles/20001106-codex-rate-card"
@@ -40,11 +44,65 @@ class CostEstimate:
     source_url: str
 
 
+def _pricing_date(value: date | datetime | int | float | str) -> date:
+    if isinstance(value, datetime):
+        if value.tzinfo is not None:
+            value = value.astimezone(timezone.utc)
+        return value.date()
+    if isinstance(value, date):
+        return value
+    if isinstance(value, (int, float)):
+        return datetime.fromtimestamp(value, timezone.utc).date()
+    return date.fromisoformat(value[:10])
+
+
+def _google_flash_rate(*, model_key: str, priced_on: date) -> ModelRate | None:
+    if model_key == "gemini-3.6-flash" and priced_on < GOOGLE_36_RELEASE_START:
+        return None
+    if model_key == "gemini-3.7-flash" and priced_on < GOOGLE_FLASH_INTRO_START:
+        return None
+    if model_key not in {"gemini-3.6-flash", "gemini-3.7-flash"}:
+        return None
+
+    model_name = "Gemini 3.7 Flash" if model_key == "gemini-3.7-flash" else "Gemini 3.6 Flash"
+    if GOOGLE_FLASH_INTRO_START <= priced_on < GOOGLE_FLASH_INTRO_END:
+        return ModelRate(
+            0.75,
+            3.75,
+            0.075,
+            f"google-{model_key}-standard-intro@2026-08-13",
+            (
+                f"Google {model_name} 프로모션 표준 단가(2026-12-31까지)"
+                " · 입력 $0.75/M · 캐시 $0.075/M · 출력 $3.75/M"
+            ),
+            GOOGLE_PRICING_URL,
+        )
+
+    effective_from = "2027-01-01" if priced_on >= GOOGLE_FLASH_INTRO_END else "2026-07-21"
+    period_label = (
+        "2027-01-01부터"
+        if priced_on >= GOOGLE_FLASH_INTRO_END
+        else "2026-07-21~2026-08-12"
+    )
+    return ModelRate(
+        1.50,
+        7.50,
+        0.15,
+        f"google-{model_key}-standard@{effective_from}",
+        (
+            f"Google {model_name} 표준 단가({period_label})"
+            " · 입력 $1.50/M · 캐시 $0.15/M · 출력 $7.50/M"
+        ),
+        GOOGLE_PRICING_URL,
+    )
+
+
 def _rate(
     *,
     provider: str,
     model: str,
     tokens_in: int,
+    priced_on: date,
 ) -> ModelRate | None:
     provider_key = provider.strip().lower()
     model_key = model.strip().lower()
@@ -56,15 +114,9 @@ def _rate(
             break
 
     if provider_key == "agy_cli":
-        if model_key == "gemini-3.6-flash":
-            return ModelRate(
-                1.50,
-                7.50,
-                0.15,
-                f"google-gemini-3.6-flash-standard@{RATE_CARD_VERSION}",
-                "Google Gemini 3.6 Flash 표준 단가 · 입력 $1.50/M · 캐시 $0.15/M · 출력 $7.50/M",
-                GOOGLE_PRICING_URL,
-            )
+        flash_rate = _google_flash_rate(model_key=model_key, priced_on=priced_on)
+        if flash_rate is not None:
+            return flash_rate
         if model_key == "gemini-3.5-flash":
             return ModelRate(
                 1.50,
@@ -182,6 +234,7 @@ def estimate_reference_cost(
     model: str,
     tokens_in: int,
     tokens_out: int,
+    priced_at: date | datetime | int | float | str,
     cached_input_tokens: int = 0,
 ) -> CostEstimate | None:
     """공식 공개 단가로 USD 상당액을 계산한다.
@@ -195,7 +248,12 @@ def estimate_reference_cost(
     if safe_input == 0 and safe_output == 0:
         return None
     cached = min(safe_input, max(0, int(cached_input_tokens or 0)))
-    rate = _rate(provider=provider, model=model, tokens_in=safe_input)
+    rate = _rate(
+        provider=provider,
+        model=model,
+        tokens_in=safe_input,
+        priced_on=_pricing_date(priced_at),
+    )
     if rate is None:
         return None
     uncached = safe_input - cached