From 31eb8c33034f2aef40c6d95b94dd8c8ad0c24650 Mon Sep 17 00:00:00 2001
From: jze9 <jze9@moon-1.home>
Date: Tue, 14 Jul 2026 15:39:47 +0500
Subject: [PATCH] =?UTF-8?q?LLM=20max=5Ftokens=201000=E2=86=928000:=20reaso?=
 =?UTF-8?q?ning-=D0=BC=D0=BE=D0=B4=D0=B5=D0=BB=D0=B8=20=D1=82=D1=80=D0=B0?=
 =?UTF-8?q?=D1=82=D0=B8=D0=BB=D0=B8=20=D0=B2=D0=B5=D1=81=D1=8C=20=D0=BB?=
 =?UTF-8?q?=D0=B8=D0=BC=D0=B8=D1=82=20=D0=BD=D0=B0=20=D1=80=D0=B0=D0=B7?=
 =?UTF-8?q?=D0=BC=D1=8B=D1=88=D0=BB=D0=B5=D0=BD=D0=B8=D1=8F=20=D0=B8=20?=
 =?UTF-8?q?=D0=B2=D0=BE=D0=B7=D0=B2=D1=80=D0=B0=D1=89=D0=B0=D0=BB=D0=B8=20?=
 =?UTF-8?q?=D0=BF=D1=83=D1=81=D1=82=D0=BE=D0=B9=20=D0=BE=D1=82=D0=B2=D0=B5?=
 =?UTF-8?q?=D1=82?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
---
 .env                  | 2 +-
 api/config.py         | 3 ++-
 api/lib/llm.py        | 3 ++-
 api/route/pipeline.py | 2 +-
 4 files changed, 6 insertions(+), 4 deletions(-)

diff --git a/.env b/.env
index 068deae..94b26f2 100644
--- a/.env
+++ b/.env
@@ -23,5 +23,5 @@ LLM_BASE_URL=
 LLM_API_KEY=
 LLM_MODEL=
 LLM_TIMEOUT=120
-LLM_MAX_TOKENS=1000
+LLM_MAX_TOKENS=8000
 LLM_TEMPERATURE=0.3
diff --git a/api/config.py b/api/config.py
index 91891e4..fb9c4d7 100644
--- a/api/config.py
+++ b/api/config.py
@@ -20,7 +20,8 @@ class Settings(BaseSettings):
     LLM_API_KEY: str = os.getenv("LLM_API_KEY", "")
     LLM_MODEL: str = os.getenv("LLM_MODEL", "")
     LLM_TIMEOUT: int = int(os.getenv("LLM_TIMEOUT", "120"))
-    LLM_MAX_TOKENS: int = int(os.getenv("LLM_MAX_TOKENS", "1000"))
+    # reasoning-модели тратят max_tokens и на размышления — 1000 им мало
+    LLM_MAX_TOKENS: int = int(os.getenv("LLM_MAX_TOKENS", "8000"))
     LLM_TEMPERATURE: float = float(os.getenv("LLM_TEMPERATURE", "0.3"))
 
     @property
diff --git a/api/lib/llm.py b/api/lib/llm.py
index ec1a674..5660e58 100644
--- a/api/lib/llm.py
+++ b/api/lib/llm.py
@@ -36,7 +36,8 @@ class LLMConfig(BaseModel):
     # before the user has picked one.
     model: str = ""
     timeout: int = 120
-    max_tokens: int = 1000
+    # у reasoning-моделей сюда входят и токены размышлений
+    max_tokens: int = 8000
     temperature: float = 0.3
     extra_headers: dict[str, str] = Field(default_factory=dict)
 
diff --git a/api/route/pipeline.py b/api/route/pipeline.py
index 6ecde63..51d1ca2 100644
--- a/api/route/pipeline.py
+++ b/api/route/pipeline.py
@@ -18,7 +18,7 @@ class LLMOverride(BaseModel):
     api_key: str = ""
     model: str
     timeout: int = 120
-    max_tokens: int = 1000
+    max_tokens: int = 8000
     temperature: float = 0.3
     extra_headers: dict[str, str] = {}
     system_prompt: str | None = None