diff --git a/README.md b/README.md
index ac00fe301..e600c9d5a 100644
--- a/README.md
+++ b/README.md
@@ -366,8 +366,8 @@ The configuration file is saved in `~/.clawcodex/config.json`. Example structure
     },
     "minimax": {
       "api_key": "your-api-key",
-      "base_url": "https://api.minimaxi.com/anthropic",
-      "default_model": "MiniMax-M2.7"
+      "base_url": "https://api.minimax.io/anthropic",
+      "default_model": "MiniMax-M3"
     },
     "openrouter": {
       "api_key": "your-api-key",
@@ -396,6 +396,14 @@ The configuration file is saved in `~/.clawcodex/config.json`. Example structure
 }
 ```
 
+The built-in Minimax provider passes an SDK base URL to the Anthropic SDK. Use
+`https://api.minimax.io/anthropic` globally or
+`https://api.minimaxi.com/anthropic` in China; the SDK appends `/v1/messages`.
+The final Messages request URLs are
+`https://api.minimax.io/anthropic/v1/messages` and
+`https://api.minimaxi.com/anthropic/v1/messages`. The OpenAI-compatible API roots are
+`https://api.minimax.io/v1` globally and `https://api.minimaxi.com/v1` in China.
+
 - **`session`** — REPL session persistence: `auto_save` writes each session automatically; `max_history` caps retained turns.
 - **`settings`** — the advisor (reviewer) feature. **`advisor_enabled` is the master switch — `false` by default, so the advisor is OFF unless you opt in** (set it to `true`, or run `/advisor <provider>:<model>` which flips it on). `advisor_provider` / `advisor_model` pick the reviewer model; `advisor_client_mode` routes the call via the client.
 - **`env`** — secrets and environment values injected at startup (e.g. `TAVILY_API_KEY` for web search). Managed via `clawcodex config`; keys here are exported into the process environment without overriding anything you already set in your shell.
@@ -660,4 +668,4 @@ If you find this useful, please **star** ⭐ the repo!
 
 ***
 
-***
\ No newline at end of file
+***
diff --git a/docs/i18n/README_ZH.md b/docs/i18n/README_ZH.md
index 6e9295291..e373c7329 100644
--- a/docs/i18n/README_ZH.md
+++ b/docs/i18n/README_ZH.md
@@ -326,8 +326,8 @@ clawcodex login
     },
     "minimax": {
       "api_key": "your-api-key",
-      "base_url": "https://api.minimaxi.com/anthropic",
-      "default_model": "MiniMax-M2.7"
+      "base_url": "https://api.minimax.io/anthropic",
+      "default_model": "MiniMax-M3"
     },
     "openrouter": {
       "api_key": "your-api-key",
@@ -355,6 +355,13 @@ clawcodex login
 }
 ```
 
+内置 Minimax provider 会把 SDK base URL 传给 Anthropic SDK：全球区域使用
+`https://api.minimax.io/anthropic`，中国区域使用
+`https://api.minimaxi.com/anthropic`，SDK 会自动追加 `/v1/messages`。最终的
+Messages 请求 URL 分别为 `https://api.minimax.io/anthropic/v1/messages` 和
+`https://api.minimaxi.com/anthropic/v1/messages`。OpenAI 兼容 API root 分别为全球区域的
+`https://api.minimax.io/v1` 和中国区域的 `https://api.minimaxi.com/v1`。
+
 - **`session`** —— REPL 会话持久化：`auto_save` 自动保存每个会话；`max_history` 限制保留的对话轮数。
 - **`settings`** —— 后台辅助功能所用的 advisor 模型（`advisor_provider` / `advisor_model`，以及控制是否经由客户端路由的 `advisor_client_mode`）。
 - **`env`** —— 启动时注入的密钥与环境变量（例如用于 Web 搜索的 `TAVILY_API_KEY`）。通过 `clawcodex config` 管理；这里的键会被导出到进程环境，但不会覆盖你在 shell 中已设置的值。
diff --git a/src/context_system/context_analyzer.py b/src/context_system/context_analyzer.py
index 9df690fa1..47fe52d85 100644
--- a/src/context_system/context_analyzer.py
+++ b/src/context_system/context_analyzer.py
@@ -36,7 +36,9 @@
     # DeepSeek V4 ships a 1M context window, like glm-5.2 (legacy deepseek-chat/
     # -reasoner are intentionally NOT matched here — see src/models/configs.py).
     "deepseek-v4": 1_000_000,
-    # Minimax defaults
+    # MiniMax model-specific windows must precede the family fallback.
+    "minimax-m3": 1_000_000,
+    "minimax-m2.7": 204_800,
     "minimax": 128_000,
     "abab": 128_000,
 }
diff --git a/src/models/configs.py b/src/models/configs.py
index 04e67352f..3eb47dd6b 100644
--- a/src/models/configs.py
+++ b/src/models/configs.py
@@ -187,6 +187,21 @@ class ModelConfig:
         max_output_tokens=8_192,
         supports_cache=True,
     ),
+    # MiniMax pricing is maintained in services/pricing.py.
+    "MiniMax-M2.7": ModelConfig(
+        model_id="MiniMax-M2.7",
+        display_name="MiniMax M2.7",
+        context_window=204_800,
+        max_output_tokens=8_192,
+        supports_cache=True,
+    ),
+    "MiniMax-M3": ModelConfig(
+        model_id="MiniMax-M3",
+        display_name="MiniMax M3",
+        context_window=1_000_000,
+        max_output_tokens=8_192,
+        supports_cache=True,
+    ),
     # Meta Muse Spark 1.1 (api.meta.ai, OpenAI-compatible). Muse Spark is a
     # server-side reasoning model (usage reports ``reasoning_tokens``); like
     # DeepSeek/GLM it exposes no Anthropic-style thinking blocks (the
diff --git a/src/providers/__init__.py b/src/providers/__init__.py
index 4dde5bd87..385caa61a 100644
--- a/src/providers/__init__.py
+++ b/src/providers/__init__.py
@@ -81,10 +81,10 @@ class ProviderInfo(TypedDict):
     },
     "minimax": {
         "label": "Minimax AI",
-        "default_base_url": "https://api.minimaxi.com/anthropic",
-        "default_model": "MiniMax-M2.7",
+        "default_base_url": "https://api.minimax.io/anthropic",
+        "default_model": "MiniMax-M3",
         "available_models": [
-            # M2 series (latest)
+            "MiniMax-M3",
             "MiniMax-M2.7",
             "MiniMax-M2.7-highspeed",
             "MiniMax-M2.5",
diff --git a/src/providers/minimax_provider.py b/src/providers/minimax_provider.py
index 03e7d477b..37a69d987 100644
--- a/src/providers/minimax_provider.py
+++ b/src/providers/minimax_provider.py
@@ -22,11 +22,11 @@ def __init__(self, *args, **kwargs):
 class MinimaxProvider(BaseProvider):
     """Minimax AI provider using Anthropic-compatible API.
 
-    Minimax provides an Anthropic-compatible endpoint at api.minimaxi.com/anthropic.
+    Minimax provides an Anthropic-compatible endpoint at api.minimax.io/anthropic.
     Uses the Anthropic SDK with Minimax-specific models.
     """
 
-    DEFAULT_BASE_URL = "https://api.minimaxi.com/anthropic"
+    DEFAULT_BASE_URL = "https://api.minimax.io/anthropic"
 
     def __init__(
         self, api_key: str, base_url: Optional[str] = None, model: Optional[str] = None
@@ -36,10 +36,10 @@ def __init__(
         Args:
             api_key: Minimax API key
             base_url: Base URL (optional, defaults to Minimax Anthropic-compatible endpoint)
-            model: Default model (default: MiniMax-M2.7)
+            model: Default model (default: MiniMax-M3)
         """
         resolved_base_url = base_url or self.DEFAULT_BASE_URL
-        super().__init__(api_key, resolved_base_url, model or "MiniMax-M2.7")
+        super().__init__(api_key, resolved_base_url, model or "MiniMax-M3")
 
         self._client_kwargs: dict[str, Any] = {"api_key": api_key}
         if resolved_base_url:
@@ -239,7 +239,7 @@ def get_available_models(self) -> list[str]:
             List of model names
         """
         return [
-            # M2 series (latest)
+            "MiniMax-M3",
             "MiniMax-M2.7",
             "MiniMax-M2.7-highspeed",
             "MiniMax-M2.5",
diff --git a/src/services/pricing.py b/src/services/pricing.py
index 90c1187a5..607c187d6 100644
--- a/src/services/pricing.py
+++ b/src/services/pricing.py
@@ -78,6 +78,20 @@
     "cache_creation": 0.435 / 1_000_000,
     "cache_read": 0.003625 / 1_000_000,
 }
+# MiniMax standard pay-as-you-go rates in USD per million tokens. The static
+# pricing table uses M3's <=512k input tier; longer requests have a higher tier.
+_TIER_MINIMAX_M3 = {
+    "input": 0.30 / 1_000_000,
+    "output": 1.20 / 1_000_000,
+    "cache_creation": 0.30 / 1_000_000,
+    "cache_read": 0.06 / 1_000_000,
+}
+_TIER_MINIMAX_M27 = {
+    "input": 0.30 / 1_000_000,
+    "output": 1.20 / 1_000_000,
+    "cache_creation": 0.375 / 1_000_000,
+    "cache_read": 0.06 / 1_000_000,
+}
 # Meta Muse Spark 1.1 (api.meta.ai, OpenAI-compatible). Meta's published rates:
 # $1.25/M input, $4.25/M output, $0.15/M cached input. OpenAI-style caching has
 # no separate cache-write charge, so ``cache_creation`` mirrors ``input``.
@@ -124,6 +138,8 @@
     # every proxied model is priced at its upstream rate.
     "deepseek-v4-flash": _TIER_DEEPSEEK_FLASH,
     "deepseek-v4-pro": _TIER_DEEPSEEK_PRO,
+    "MiniMax-M3": _TIER_MINIMAX_M3,
+    "MiniMax-M2.7": _TIER_MINIMAX_M27,
     # Meta Muse Spark (api.meta.ai)
     "muse-spark-1.1": _TIER_MUSE_SPARK,
 }
diff --git a/tests/test_context_window_1m.py b/tests/test_context_window_1m.py
index 34690b6bd..c4ee74e32 100644
--- a/tests/test_context_window_1m.py
+++ b/tests/test_context_window_1m.py
@@ -27,6 +27,13 @@ def test_glm_4_legacy_not_promoted_to_1m():
     assert canonical_window("glm-4") == 128_000
 
 
+def test_minimax_current_context_windows():
+    assert display_window("MiniMax-M3") == 1_000_000
+    assert canonical_window("MiniMax-M3") == 1_000_000
+    assert display_window("MiniMax-M2.7") == 204_800
+    assert canonical_window("MiniMax-M2.7") == 204_800
+
+
 def test_legacy_deepseek_chat_not_promoted_to_1m():
     # Only deepseek-v4* gets 1M; legacy deepseek-chat/-reasoner do not.
     assert display_window("deepseek-chat") != 1_000_000
diff --git a/tests/test_pricing_status_bar.py b/tests/test_pricing_status_bar.py
index 232163464..15f6f0b65 100644
--- a/tests/test_pricing_status_bar.py
+++ b/tests/test_pricing_status_bar.py
@@ -35,6 +35,18 @@ def test_exact_match_sonnet_4_6(self) -> None:
         self.assertEqual(p["input"], 3.0 / 1_000_000)
         self.assertEqual(p["output"], 15.0 / 1_000_000)
 
+    def test_exact_match_minimax_models(self) -> None:
+        m3 = get_pricing("MiniMax-M3")
+        self.assertEqual(m3["input"], 0.30 / 1_000_000)
+        self.assertEqual(m3["output"], 1.20 / 1_000_000)
+        self.assertEqual(m3["cache_read"], 0.06 / 1_000_000)
+
+        m27 = get_pricing("MiniMax-M2.7")
+        self.assertEqual(m27["input"], 0.30 / 1_000_000)
+        self.assertEqual(m27["output"], 1.20 / 1_000_000)
+        self.assertEqual(m27["cache_creation"], 0.375 / 1_000_000)
+        self.assertEqual(m27["cache_read"], 0.06 / 1_000_000)
+
     def test_family_prefix_falls_back_for_future_opus_variant(self) -> None:
         # A model name not in the exact table but matching the
         # ``claude-opus-4-7`` family prefix → 5/25 tier.
diff --git a/tests/test_provider_registry.py b/tests/test_provider_registry.py
index 4dcf421f4..36e694823 100644
--- a/tests/test_provider_registry.py
+++ b/tests/test_provider_registry.py
@@ -20,6 +20,7 @@
     resolve_api_key,
 )
 from src.providers.base import ChatMessage
+from src.providers.minimax_provider import MinimaxProvider
 from src.providers.openai_compatible import OpenAICompatibleProvider
 from src.providers.openai_compatible_specs import (
     SPECS_BY_ID,
@@ -86,6 +87,17 @@ def test_default_model_is_in_available_models(self):
         for spec in SPECS_BY_ID.values():
             self.assertIn(spec.default_model, spec.available_models, spec.id)
 
+    def test_minimax_current_defaults_and_models(self):
+        info = PROVIDER_INFO["minimax"]
+        self.assertEqual(info["default_base_url"], "https://api.minimax.io/anthropic")
+        self.assertEqual(info["default_model"], "MiniMax-M3")
+        self.assertEqual(info["available_models"][:2], ["MiniMax-M3", "MiniMax-M2.7"])
+
+        provider = MinimaxProvider(api_key="k")
+        self.assertEqual(provider.base_url, "https://api.minimax.io/anthropic")
+        self.assertEqual(provider.model, "MiniMax-M3")
+        self.assertEqual(provider.get_available_models()[:2], ["MiniMax-M3", "MiniMax-M2.7"])
+
     def test_hand_written_providers_not_shadowed(self):
         # The registry must hold only the *new* providers — never override the
         # bespoke hand-written ones (deepseek cache logic, zai GLM aliasing …).