diff --git a/.env.example b/.env.example
index 77fef51..5983188 100644
--- a/.env.example
+++ b/.env.example
@@ -21,9 +21,9 @@ LLAMACPP_BASE_URL="http://localhost:8080/v1"
 # All Claude model requests are mapped to these models, plain model is fallback
 # Format: provider_type/model/name
 # Valid providers: "nvidia_nim" | "open_router" | "deepseek" | "lmstudio" | "llamacpp"
-MODEL_OPUS="nvidia_nim/z-ai/glm4.7"
-MODEL_SONNET="open_router/arcee-ai/trinity-large-preview:free"
-MODEL_HAIKU="open_router/stepfun/step-3.5-flash:free"
+MODEL_OPUS=
+MODEL_SONNET=
+MODEL_HAIKU=
 MODEL="nvidia_nim/z-ai/glm4.7"
 
 
diff --git a/README.md b/README.md
index 6a2e11d..2309e61 100644
--- a/README.md
+++ b/README.md
@@ -76,9 +76,9 @@ Choose your provider and edit `.env`:
 ```dotenv
 NVIDIA_NIM_API_KEY="nvapi-your-key-here"
 
-MODEL_OPUS="nvidia_nim/z-ai/glm4.7"
-MODEL_SONNET="nvidia_nim/moonshotai/kimi-k2-thinking"
-MODEL_HAIKU="nvidia_nim/stepfun-ai/step-3.5-flash"
+MODEL_OPUS=
+MODEL_SONNET=
+MODEL_HAIKU=
 MODEL="nvidia_nim/z-ai/glm4.7"                     # fallback
 
 # Global switch for provider reasoning requests and Claude thinking blocks.
@@ -502,10 +502,10 @@ Configure via `WHISPER_DEVICE` (`cpu` | `cuda` | `nvidia_nim`) and `WHISPER_MODE
 
 | Variable             | Description                                                           | Default                                           |
 | -------------------- | --------------------------------------------------------------------- | ------------------------------------------------- |
-| `MODEL`              | Fallback model (`provider/model/name` format; invalid prefix → error) | `nvidia_nim/stepfun-ai/step-3.5-flash`            |
-| `MODEL_OPUS`         | Model for Claude Opus requests (falls back to `MODEL`)                | `nvidia_nim/z-ai/glm4.7`                          |
-| `MODEL_SONNET`       | Model for Claude Sonnet requests (falls back to `MODEL`)              | `open_router/arcee-ai/trinity-large-preview:free` |
-| `MODEL_HAIKU`        | Model for Claude Haiku requests (falls back to `MODEL`)               | `open_router/stepfun/step-3.5-flash:free`         |
+| `MODEL`              | Fallback model (`provider/model/name` format; invalid prefix → error) | `nvidia_nim/z-ai/glm4.7`                          |
+| `MODEL_OPUS`         | Model for Claude Opus requests; empty falls back to `MODEL`           | empty                                             |
+| `MODEL_SONNET`       | Model for Claude Sonnet requests; empty falls back to `MODEL`         | empty                                             |
+| `MODEL_HAIKU`        | Model for Claude Haiku requests; empty falls back to `MODEL`          | empty                                             |
 | `NVIDIA_NIM_API_KEY`    | NVIDIA API key                                                        | required for NIM                                  |
 | `ENABLE_THINKING`    | Global switch for provider reasoning requests and Claude thinking blocks. Set `false` to hide thinking across all providers. | `true` |
 | `OPENROUTER_API_KEY` | OpenRouter API key                                                    | required for OpenRouter                           |
diff --git a/config/env.example b/config/env.example
index a77fd4e..6b0faa5 100644
--- a/config/env.example
+++ b/config/env.example
@@ -17,9 +17,9 @@ LM_STUDIO_BASE_URL="http://localhost:1234/v1"
 # All Claude model requests are mapped to these models, plain model is fallback
 # Format: provider_type/model/name
 # Valid providers: "nvidia_nim" | "open_router" | "deepseek" | "lmstudio" | "llamacpp"
-MODEL_OPUS="nvidia_nim/z-ai/glm4.7"
-MODEL_SONNET="open_router/arcee-ai/trinity-large-preview:free"
-MODEL_HAIKU="open_router/stepfun/step-3.5-flash:free"
+MODEL_OPUS=
+MODEL_SONNET=
+MODEL_HAIKU=
 MODEL="nvidia_nim/z-ai/glm4.7"
 
 
diff --git a/config/settings.py b/config/settings.py
index deb6888..59c69b0 100644
--- a/config/settings.py
+++ b/config/settings.py
@@ -119,7 +119,7 @@ class Settings(BaseSettings):
     # ==================== Model ====================
     # All Claude model requests are mapped to this single model (fallback)
     # Format: provider_type/model/name
-    model: str = "nvidia_nim/stepfun-ai/step-3.5-flash"
+    model: str = "nvidia_nim/z-ai/glm4.7"
 
     # Per-model overrides (optional, falls back to MODEL)
     # Each can use a different provider
@@ -217,6 +217,9 @@ class Settings(BaseSettings):
         "allowed_telegram_user_id",
         "discord_bot_token",
         "allowed_discord_channels",
+        "model_opus",
+        "model_sonnet",
+        "model_haiku",
         mode="before",
     )
     @classmethod
diff --git a/tests/config/test_config.py b/tests/config/test_config.py
index ba53c0d..ae64bec 100644
--- a/tests/config/test_config.py
+++ b/tests/config/test_config.py
@@ -24,7 +24,7 @@ class TestSettings:
         monkeypatch.delenv("HTTP_READ_TIMEOUT", raising=False)
         monkeypatch.setitem(Settings.model_config, "env_file", ())
         settings = Settings()
-        assert settings.model == "nvidia_nim/stepfun-ai/step-3.5-flash"
+        assert settings.model == "nvidia_nim/z-ai/glm4.7"
         assert isinstance(settings.provider_rate_limit, int)
         assert isinstance(settings.provider_rate_window, int)
         assert isinstance(settings.nim.temperature, float)
@@ -391,6 +391,19 @@ class TestPerModelMapping:
         s = Settings()
         assert s.model_opus == "open_router/deepseek/deepseek-r1"
 
+    @pytest.mark.parametrize("env_var", ["MODEL_OPUS", "MODEL_SONNET", "MODEL_HAIKU"])
+    def test_empty_model_override_env_is_unset(self, monkeypatch, env_var):
+        """Empty per-model override env vars are treated as unset."""
+        from config.settings import Settings
+
+        monkeypatch.setenv(env_var, "")
+        s = Settings()
+        assert getattr(s, env_var.lower()) is None
+        assert (
+            s.resolve_model(f"claude-{env_var.removeprefix('MODEL_').lower()}-4")
+            == s.model
+        )
+
     @pytest.mark.parametrize(
         "env_vars,expected_model,expected_haiku",
         [