diff --git a/README.md b/README.md
index 018d432..5606d53 100644
--- a/README.md
+++ b/README.md
@@ -256,6 +256,23 @@ uncommon-route provider add google     AIza...
 uncommon-route serve
 ```
 
+MiniMax uses the global OpenAI-compatible endpoint by default. Pass the China
+endpoint explicitly when needed:
+
+```bash
+uncommon-route provider add minimax eyJ-...
+uncommon-route provider add minimax eyJ-... --url https://api.minimaxi.com/v1
+```
+
+| Region | OpenAI-compatible base URL | Anthropic-compatible base URL |
+|---|---|---|
+| Global | `https://api.minimax.io/v1` | `https://api.minimax.io/anthropic` |
+| China | `https://api.minimaxi.com/v1` | `https://api.minimaxi.com/anthropic` |
+
+Provider configuration stores the OpenAI-compatible base URL. Anthropic
+requests use the corresponding `/anthropic` base and append `/v1/messages`
+internally.
+
 > UncommonRoute doesn't automatically read `OPENAI_API_KEY` or `ANTHROPIC_API_KEY`. Use `init`, a saved connection, or one of the manual setup paths above.
 
 ### Routing Modes
diff --git a/tests/test_anthropic_compat.py b/tests/test_anthropic_compat.py
index a1d462d..519b1ee 100644
--- a/tests/test_anthropic_compat.py
+++ b/tests/test_anthropic_compat.py
@@ -1931,9 +1931,18 @@ def handler(request: httpx.Request) -> httpx.Response:
         finally:
             asyncio.run(async_client.aclose())
 
+    @pytest.mark.parametrize(
+        ("base_url", "expected_url"),
+        [
+            ("https://api.minimax.io/v1", "https://api.minimax.io/anthropic/v1/messages"),
+            ("https://api.minimaxi.com/v1", "https://api.minimaxi.com/anthropic/v1/messages"),
+        ],
+    )
     def test_messages_use_minimax_anthropic_endpoint_for_direct_provider(
         self,
         monkeypatch: pytest.MonkeyPatch,
+        base_url: str,
+        expected_url: str,
     ) -> None:
         captured: dict[str, object] = {}
 
@@ -1947,7 +1956,7 @@ def handler(request: httpx.Request) -> httpx.Response:
                     "id": "msg_minimax_byok",
                     "type": "message",
                     "role": "assistant",
-                    "model": "minimax/minimax-m2.1",
+                    "model": "MiniMax-M3",
                     "content": [{"type": "text", "text": "ok"}],
                     "stop_reason": "end_turn",
                     "stop_sequence": None,
@@ -1964,32 +1973,44 @@ def handler(request: httpx.Request) -> httpx.Response:
                 "minimax": ProviderEntry(
                     name="minimax",
                     api_key="mm-key-123",
-                    base_url="https://api.minimax.io/v1",
-                    models=["minimax/minimax-m2.1"],
+                    base_url=base_url,
+                    models=["minimax/minimax-m3"],
                 ),
             })
             app = create_app(
-                upstream="https://api.commonstack.ai/v1",
+                upstream="https://primary.example/v1",
                 providers_config=providers,
             )
             client = TestClient(app, raise_server_exceptions=False)
             resp = client.post("/v1/messages", json={
-                "model": "minimax/minimax-m2.1",
+                "model": "minimax/minimax-m3",
                 "max_tokens": 32,
                 "messages": [{"role": "user", "content": "hello"}],
             })
 
             assert resp.status_code == 200
-            assert captured["url"] == "https://api.minimax.io/anthropic/v1/messages"
+            assert captured["url"] == expected_url
             headers = captured["headers"]
             assert isinstance(headers, dict)
             assert headers["x-api-key"] == "mm-key-123"
+            body = captured["body"]
+            assert isinstance(body, dict)
+            assert body["model"] == "MiniMax-M3"
         finally:
             asyncio.run(async_client.aclose())
 
-    def test_chat_completions_keep_openai_transport_for_minimax_models(
+    @pytest.mark.parametrize(
+        ("base_url", "expected_url"),
+        [
+            ("https://api.minimax.io/v1", "https://api.minimax.io/v1/chat/completions"),
+            ("https://api.minimaxi.com/v1", "https://api.minimaxi.com/v1/chat/completions"),
+        ],
+    )
+    def test_chat_completions_use_minimax_openai_endpoint_for_direct_provider(
         self,
         monkeypatch: pytest.MonkeyPatch,
+        base_url: str,
+        expected_url: str,
     ) -> None:
         captured: dict[str, object] = {}
 
@@ -2003,7 +2024,7 @@ def handler(request: httpx.Request) -> httpx.Response:
                     "id": "chatcmpl_minimax",
                     "object": "chat.completion",
                     "created": 1,
-                    "model": "minimax/minimax-m2.1",
+                    "model": "MiniMax-M3",
                     "choices": [{
                         "index": 0,
                         "message": {"role": "assistant", "content": "done"},
@@ -2016,13 +2037,23 @@ def handler(request: httpx.Request) -> httpx.Response:
 
         async_client = httpx.AsyncClient(transport=httpx.MockTransport(handler))
         monkeypatch.setattr("uncommon_route.proxy._get_client", lambda: async_client)
-        monkeypatch.setenv("UNCOMMON_ROUTE_API_KEY", "env-key-123")
 
         try:
-            app = create_app(upstream="https://api.commonstack.ai/v1")
+            providers = ProvidersConfig(providers={
+                "minimax": ProviderEntry(
+                    name="minimax",
+                    api_key="mm-key-123",
+                    base_url=base_url,
+                    models=["minimax/minimax-m3"],
+                ),
+            })
+            app = create_app(
+                upstream="https://primary.example/v1",
+                providers_config=providers,
+            )
             client = TestClient(app, raise_server_exceptions=False)
             resp = client.post("/v1/chat/completions", json={
-                "model": "minimax/minimax-m2.1",
+                "model": "minimax/minimax-m3",
                 "max_tokens": 32,
                 "messages": [{"role": "user", "content": "hello"}],
             })
@@ -2031,10 +2062,13 @@ def handler(request: httpx.Request) -> httpx.Response:
             assert resp.headers["x-uncommon-route-requested-transport"] == "openai-chat"
             assert resp.headers["x-uncommon-route-transport"] == "openai-chat"
             assert resp.headers["x-uncommon-route-transport-source"] == "ingress-policy"
-            assert captured["url"] == "https://api.commonstack.ai/v1/chat/completions"
+            assert captured["url"] == expected_url
+            headers = captured["headers"]
+            assert isinstance(headers, dict)
+            assert headers["authorization"] == "Bearer mm-key-123"
             body = captured["body"]
             assert isinstance(body, dict)
-            assert body["model"] == "minimax/minimax-m2.1"
+            assert body["model"] == "MiniMax-M3"
         finally:
             asyncio.run(async_client.aclose())
 
diff --git a/tests/test_providers.py b/tests/test_providers.py
index 352f7e2..528a840 100644
--- a/tests/test_providers.py
+++ b/tests/test_providers.py
@@ -11,6 +11,7 @@
     ProviderEntry,
     add_provider,
     load_providers,
+    resolve_upstream_model,
     remove_provider,
     select_preferred_model,
 )
@@ -41,12 +42,21 @@ def test_add_provider_with_plan(self) -> None:
         cfg = add_provider("minimax", "eyJ-test", plan="coding-plan")
         entry = cfg.providers["minimax"]
         assert entry.plan == "coding-plan"
-        assert "minimax/minimax-m2.5" in entry.models
+        assert entry.models == [
+            "minimax/minimax-m2.5",
+            "minimax/minimax-m3",
+            "minimax/minimax-m2.7",
+        ]
 
     def test_add_provider_custom_url(self) -> None:
         cfg = add_provider("openai", "sk-openai", base_url="https://my-proxy.com/v1")
         assert cfg.providers["openai"].base_url == "https://my-proxy.com/v1"
 
+    def test_resolve_minimax_upstream_model_ids(self) -> None:
+        assert resolve_upstream_model("minimax", "minimax/minimax-m3") == "MiniMax-M3"
+        assert resolve_upstream_model("minimax", "minimax/minimax-m2.7") == "MiniMax-M2.7"
+        assert resolve_upstream_model("minimax", "minimax/unknown") == "minimax/unknown"
+
     def test_add_provider_custom_models(self) -> None:
         cfg = add_provider(
             "custom",
diff --git a/tests/test_proxy.py b/tests/test_proxy.py
index 08427f0..1b7043b 100644
--- a/tests/test_proxy.py
+++ b/tests/test_proxy.py
@@ -17,6 +17,7 @@
 from uncommon_route.model_experience import InMemoryModelExperienceStorage, ModelExperienceStore
 from uncommon_route.providers import ProviderEntry, ProvidersConfig
 from uncommon_route.proxy import (
+    UpstreamSemanticCompressor,
     _extract_current_message,
     _extract_prompt,
     _normalize_reasoning_content_chunk,
@@ -77,6 +78,36 @@ def _build_test_mapper(*model_ids: str) -> ModelMapper:
     return mapper
 
 
+class TestUpstreamSemanticCompressor:
+    def test_direct_minimax_provider_uses_upstream_model_id(self) -> None:
+        providers = ProvidersConfig(providers={
+            "minimax": ProviderEntry(
+                name="minimax",
+                api_key="test-key",
+                base_url="https://api.minimax.io/v1",
+                models=["minimax/minimax-m3"],
+            ),
+        })
+        compressor = UpstreamSemanticCompressor(
+            upstream_chat="https://primary.example/v1/chat/completions",
+            primary_api_key="primary-test-key",
+            providers_config=providers,
+            model_mapper=ModelMapper("https://primary.example/v1"),
+            composition_policy=CompositionPolicy(),
+        )
+
+        resolved = compressor._resolve_request(
+            "minimax/minimax-m3",
+            httpx.Request("POST", "https://local.example/v1/chat/completions"),
+        )
+
+        assert resolved is not None
+        target_url, headers, upstream_model = resolved
+        assert target_url == "https://api.minimax.io/v1/chat/completions"
+        assert headers["authorization"] == "Bearer test-key"
+        assert upstream_model == "MiniMax-M3"
+
+
 class TestPromptExtraction:
     def test_extract_prompt_ignores_claude_code_wrapper_blocks(self) -> None:
         prompt, system_prompt, max_tokens = _extract_prompt({
diff --git a/uncommon_route/providers.py b/uncommon_route/providers.py
index 338c3de..8adfccd 100644
--- a/uncommon_route/providers.py
+++ b/uncommon_route/providers.py
@@ -12,7 +12,11 @@
         "minimax": {
           "api_key": "eyJ...",
           "base_url": "https://api.minimax.io/v1",
-          "models": ["minimax/minimax-m2.5"],
+          "models": [
+            "minimax/minimax-m2.5",
+            "minimax/minimax-m3",
+            "minimax/minimax-m2.7"
+          ],
           "plan": "coding-plan"
         },
         "deepseek": {
@@ -48,7 +52,7 @@
 }
 
 PROVIDER_MODELS: dict[str, list[str]] = {
-    "minimax": ["minimax/minimax-m2.5"],
+    "minimax": ["minimax/minimax-m2.5", "minimax/minimax-m3", "minimax/minimax-m2.7"],
     "deepseek": ["deepseek/deepseek-chat", "deepseek/deepseek-reasoner"],
     "openai": ["openai/gpt-4o-mini", "openai/gpt-4o", "openai/gpt-5.2", "openai/gpt-5.2-codex", "openai/o1-mini", "openai/o3", "openai/o4-mini"],
     "anthropic": ["anthropic/claude-haiku-4.5", "anthropic/claude-sonnet-4.6", "anthropic/claude-opus-4.6"],
@@ -57,6 +61,17 @@
     "moonshot": ["moonshot/kimi-k2.5"],
 }
 
+UPSTREAM_MODEL_IDS: dict[str, dict[str, str]] = {
+    "minimax": {
+        "minimax/minimax-m3": "MiniMax-M3",
+        "minimax/minimax-m2.7": "MiniMax-M2.7",
+    },
+}
+
+
+def resolve_upstream_model(provider_name: str, model_id: str) -> str:
+    return UPSTREAM_MODEL_IDS.get(provider_name, {}).get(model_id, model_id)
+
 
 @dataclass
 class ProviderEntry:
diff --git a/uncommon_route/proxy.py b/uncommon_route/proxy.py
index fc341aa..de6f775 100644
--- a/uncommon_route/proxy.py
+++ b/uncommon_route/proxy.py
@@ -106,6 +106,7 @@
     ProvidersConfig,
     add_provider,
     load_providers,
+    resolve_upstream_model,
     remove_provider,
     verify_key,
 )
@@ -914,7 +915,7 @@ def _resolve_request(self, model_id: str, request: Request) -> tuple[str, dict[s
         provider_entry = self._providers.get_for_model(model_id)
         if provider_entry and provider_entry.base_url:
             target_chat_url = f"{provider_entry.base_url.rstrip('/')}/chat/completions"
-            upstream_model = model_id
+            upstream_model = resolve_upstream_model(provider_entry.name, model_id)
         elif self._upstream_chat:
             target_chat_url = self._upstream_chat
             upstream_model = self._mapper.resolve(model_id)
@@ -4635,7 +4636,9 @@ def _prepare_attempt(model_name: str) -> dict[str, Any]:
                     attempt_headers[_ORIGINAL_MODEL_HEADER] = requested_model
 
             resolved_model = model_name
-            if not attempt_provider_entry:
+            if attempt_provider_entry:
+                resolved_model = resolve_upstream_model(attempt_provider_entry.name, model_name)
+            else:
                 resolved_model = _mapper.resolve(model_name)
             attempt_upstream_body["model"] = resolved_model