diff --git a/.github/copilot-instructions.md b/.github/copilot-instructions.md
index bcf95a7..b7d8b51 100644
--- a/.github/copilot-instructions.md
+++ b/.github/copilot-instructions.md
@@ -5,7 +5,7 @@
 - No manual virtualenv activation is needed when using Makefile targets (`make check`, `make test`, etc.); the Makefile handles it.
 - Only activate `.venv` manually for direct Python/pip commands run outside Make targets.
 - Python 3.10+, all dependencies installed in `.venv`
-- Gemini model: `gemini-3.1-flash-lite-preview` via `google-genai` package (NOT the deprecated `google.generativeai`)
+- Gemini model: `gemini-3.1-flash-lite` via `google-genai` package (NOT the deprecated `google.generativeai`)
 
 ## After every code change
 
diff --git a/AGENTS.md b/AGENTS.md
index 2e2ad75..299e6eb 100644
--- a/AGENTS.md
+++ b/AGENTS.md
@@ -3,7 +3,7 @@
 This document defines the persona, context, and instruction sets for the AI agents used in Immermatch.
 
 **LLM Provider:** Google AI Studio (Gemini)
-**Model:** gemini-3.1-flash-lite-preview
+**Model:** gemini-3.1-flash-lite
 **Package:** `google-genai` (not the deprecated `google.generativeai`)
 
 ## Environment Guidance (All Agents)
@@ -208,7 +208,7 @@ class SearchProvider(Protocol):
 
 ```python
 # Gemini model and retry
-MODEL = "gemini-3.1-flash-lite-preview"
+MODEL = "gemini-3.1-flash-lite"
 MAX_RETRIES = 5
 BASE_DELAY = 3  # seconds, exponential backoff with jitter
 
diff --git a/CLAUDE.md b/CLAUDE.md
index d95bea4..ad91d8a 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -8,7 +8,7 @@
 ```
 
 - Python 3.10+, all dependencies in `.venv`
-- Gemini model: `gemini-3.1-flash-lite-preview` via `google-genai` (NOT `google.generativeai`)
+- Gemini model: `gemini-3.1-flash-lite` via `google-genai` (NOT `google.generativeai`)
 
 ## After every code change — run automatically, don't ask
 
diff --git a/immermatch/llm.py b/immermatch/llm.py
index 5d2c874..21e0de8 100644
--- a/immermatch/llm.py
+++ b/immermatch/llm.py
@@ -15,7 +15,7 @@
 MAX_RETRIES = 5
 BASE_DELAY = 3  # seconds
 
-MODEL = "gemini-3.1-flash-lite-preview"
+MODEL = "gemini-3.1-flash-lite"
 
 # Concurrency limiter — prevents thundering-herd 429s when many threads
 # call Gemini simultaneously (e.g. parallel job evaluation).