diff --git a/cmd/moonbridge/minimax_registry_test.go b/cmd/moonbridge/minimax_registry_test.go
new file mode 100644
index 00000000..fd7aeb45
--- /dev/null
+++ b/cmd/moonbridge/minimax_registry_test.go
@@ -0,0 +1,136 @@
+package main
+
+import (
+	"encoding/json"
+	"os"
+	"path/filepath"
+	"slices"
+	"strings"
+	"testing"
+
+	"gopkg.in/yaml.v3"
+	"moonbridge/internal/config"
+)
+
+func TestExampleConfigIncludesMiniMaxRegistry(t *testing.T) {
+	repoRoot := filepath.Join("..", "..")
+	example, err := os.ReadFile(filepath.Join(repoRoot, "config.example.yml"))
+	if err != nil {
+		t.Fatalf("read example config: %v", err)
+	}
+	var fileConfig config.FileConfig
+	if err := yaml.Unmarshal(example, &fileConfig); err != nil {
+		t.Fatalf("decode example config: %v", err)
+	}
+	for _, modelID := range []string{"MiniMax-M3", "MiniMax-M2.7"} {
+		if _, ok := fileConfig.Models[modelID]; !ok {
+			t.Fatalf("model %q is missing", modelID)
+		}
+	}
+
+	providerTests := []struct {
+		key      string
+		protocol string
+		baseURL  string
+	}{
+		{"minimax-global-anthropic", config.ProtocolAnthropic, "https://api.minimax.io/anthropic"},
+		{"minimax-global-openai", config.ProtocolOpenAIChat, "https://api.minimax.io/v1"},
+		{"minimax-cn-anthropic", config.ProtocolAnthropic, "https://api.minimaxi.com/anthropic"},
+		{"minimax-cn-openai", config.ProtocolOpenAIChat, "https://api.minimaxi.com/v1"},
+	}
+	for _, tt := range providerTests {
+		t.Run(tt.key, func(t *testing.T) {
+			provider, ok := fileConfig.Providers[tt.key]
+			if !ok {
+				t.Fatalf("provider %q is missing", tt.key)
+			}
+			if provider.Protocol != tt.protocol {
+				t.Fatalf("protocol = %q, want %q", provider.Protocol, tt.protocol)
+			}
+			if provider.BaseURL != tt.baseURL {
+				t.Fatalf("base URL = %q, want %q", provider.BaseURL, tt.baseURL)
+			}
+			assertMiniMaxOffers(t, provider.Offers)
+		})
+	}
+
+	for _, docsRoot := range []string{
+		"https://platform.minimax.io/docs",
+		"https://platform.minimaxi.com/docs",
+	} {
+		if !strings.Contains(string(example), docsRoot) {
+			t.Fatalf("example config is missing documentation root %q", docsRoot)
+		}
+	}
+}
+
+func TestMiniMaxModelMetadata(t *testing.T) {
+	tests := []struct {
+		modelID       string
+		contextWindow int
+		modalities    []string
+		reasoning     []string
+	}{
+		{"MiniMax-M3", 1000000, []string{"text", "image", "video"}, []string{"adaptive", "disabled"}},
+		{"MiniMax-M2.7", 204800, []string{"text"}, []string{"always_on"}},
+	}
+
+	for _, tt := range tests {
+		t.Run(tt.modelID, func(t *testing.T) {
+			path := filepath.Join("..", "..", "metadata", "models", tt.modelID+".json")
+			raw, err := os.ReadFile(path)
+			if err != nil {
+				t.Fatalf("read model metadata: %v", err)
+			}
+			var metadata config.ModelDefFileConfig
+			if err := json.Unmarshal(raw, &metadata); err != nil {
+				t.Fatalf("decode model metadata: %v", err)
+			}
+			if metadata.ContextWindow != tt.contextWindow {
+				t.Fatalf("context window = %d, want %d", metadata.ContextWindow, tt.contextWindow)
+			}
+			if metadata.SupportsReasoning == nil || !*metadata.SupportsReasoning {
+				t.Fatal("supports_reasoning = false, want true")
+			}
+			gotReasoning := make([]string, 0, len(metadata.SupportedReasoningLevels))
+			for _, level := range metadata.SupportedReasoningLevels {
+				gotReasoning = append(gotReasoning, level.Effort)
+			}
+			if !slices.Equal(gotReasoning, tt.reasoning) {
+				t.Fatalf("supported reasoning levels = %v, want %v", gotReasoning, tt.reasoning)
+			}
+			if !slices.Equal(metadata.InputModalities, tt.modalities) {
+				t.Fatalf("input modalities = %v, want %v", metadata.InputModalities, tt.modalities)
+			}
+		})
+	}
+}
+
+func assertMiniMaxOffers(t *testing.T, offers []config.OfferFileConfig) {
+	t.Helper()
+	want := map[string]config.ModelPricingFileConfig{
+		"MiniMax-M3": {
+			InputPrice:     0.6,
+			OutputPrice:    2.4,
+			CacheReadPrice: 0.12,
+		},
+		"MiniMax-M2.7": {
+			InputPrice:      0.3,
+			OutputPrice:     1.2,
+			CacheWritePrice: 0.375,
+			CacheReadPrice:  0.06,
+		},
+	}
+	if len(offers) != len(want) {
+		t.Fatalf("offer count = %d, want %d", len(offers), len(want))
+	}
+	for _, offer := range offers {
+		pricing, ok := want[offer.Model]
+		if !ok {
+			t.Fatalf("unexpected model offer %q", offer.Model)
+		}
+		if offer.Pricing != pricing {
+			t.Fatalf("pricing for %q = %+v, want %+v", offer.Model, offer.Pricing, pricing)
+		}
+	}
+}
diff --git a/config.example.yml b/config.example.yml
index 118ccacd..6bee0ee2 100644
--- a/config.example.yml
+++ b/config.example.yml
@@ -158,6 +158,12 @@ models:
   claude-sonnet-4-20250514:
     extensions: {}   # no overrides, metadata only
 
+  MiniMax-M3:
+    extensions: {}   # no overrides, metadata only
+
+  MiniMax-M2.7:
+    extensions: {}   # no overrides, metadata only
+
   # kimi-for-coding:
   #   context_window: 128000
   #   extensions:
@@ -229,6 +235,82 @@ providers:
           cache_write_price: 3.75
           cache_read_price: 0.30
 
+  # MiniMax global endpoints: https://platform.minimax.io/docs
+  minimax-global-anthropic:
+    protocol: "anthropic"
+    base_url: "https://api.minimax.io/anthropic"
+    api_key: "replace-me"
+    version: "2023-06-01"
+    user_agent: "moonbridge/1.0"
+    offers:
+      - model: MiniMax-M3
+        pricing:
+          input_price: 0.6
+          output_price: 2.4
+          cache_read_price: 0.12
+      - model: MiniMax-M2.7
+        pricing:
+          input_price: 0.3
+          output_price: 1.2
+          cache_write_price: 0.375
+          cache_read_price: 0.06
+
+  minimax-global-openai:
+    protocol: "openai-chat"
+    base_url: "https://api.minimax.io/v1"
+    api_key: "replace-me"
+    user_agent: "moonbridge/1.0"
+    offers:
+      - model: MiniMax-M3
+        pricing:
+          input_price: 0.6
+          output_price: 2.4
+          cache_read_price: 0.12
+      - model: MiniMax-M2.7
+        pricing:
+          input_price: 0.3
+          output_price: 1.2
+          cache_write_price: 0.375
+          cache_read_price: 0.06
+
+  # MiniMax China endpoints: https://platform.minimaxi.com/docs
+  minimax-cn-anthropic:
+    protocol: "anthropic"
+    base_url: "https://api.minimaxi.com/anthropic"
+    api_key: "replace-me"
+    version: "2023-06-01"
+    user_agent: "moonbridge/1.0"
+    offers:
+      - model: MiniMax-M3
+        pricing:
+          input_price: 0.6
+          output_price: 2.4
+          cache_read_price: 0.12
+      - model: MiniMax-M2.7
+        pricing:
+          input_price: 0.3
+          output_price: 1.2
+          cache_write_price: 0.375
+          cache_read_price: 0.06
+
+  minimax-cn-openai:
+    protocol: "openai-chat"
+    base_url: "https://api.minimaxi.com/v1"
+    api_key: "replace-me"
+    user_agent: "moonbridge/1.0"
+    offers:
+      - model: MiniMax-M3
+        pricing:
+          input_price: 0.6
+          output_price: 2.4
+          cache_read_price: 0.12
+      - model: MiniMax-M2.7
+        pricing:
+          input_price: 0.3
+          output_price: 1.2
+          cache_write_price: 0.375
+          cache_read_price: 0.06
+
   # Google Gemini (Vertex AI via GenAI SDK)
   # gemini:
   #   protocol: "google-genai"
diff --git a/internal/protocol/anthropic/client_test.go b/internal/protocol/anthropic/client_test.go
index 70c3fb4d..ae47ad8d 100644
--- a/internal/protocol/anthropic/client_test.go
+++ b/internal/protocol/anthropic/client_test.go
@@ -5,6 +5,7 @@ import (
 	"encoding/json"
 	"io"
 	"net/http"
+	"net/http/httptest"
 	"strings"
 	"testing"
 
@@ -77,6 +78,39 @@ func TestClientCreateMessageSendsHeadersAndParsesResponse(t *testing.T) {
 	}
 }
 
+func TestClientCreateMessagePreservesBaseURLPathPrefix(t *testing.T) {
+	var requestPath string
+	server := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
+		requestPath = r.URL.Path
+		w.Header().Set("content-type", "application/json")
+		_ = json.NewEncoder(w).Encode(anthropic.MessageResponse{
+			ID:      "msg_123",
+			Type:    "message",
+			Role:    "assistant",
+			Content: []anthropic.ContentBlock{{Type: "text", Text: "Hello"}},
+		})
+	}))
+	defer server.Close()
+
+	client := anthropic.NewClient(anthropic.ClientConfig{
+		BaseURL: server.URL + "/anthropic",
+		APIKey:  "upstream-key",
+		Version: "2023-06-01",
+		Client:  server.Client(),
+	})
+	_, err := client.CreateMessage(context.Background(), anthropic.MessageRequest{
+		Model:     "test-model",
+		MaxTokens: 64,
+		Messages:  []anthropic.Message{{Role: "user", Content: []anthropic.ContentBlock{{Type: "text", Text: "Hi"}}}},
+	})
+	if err != nil {
+		t.Fatalf("CreateMessage() error = %v", err)
+	}
+	if requestPath != "/anthropic/v1/messages" {
+		t.Fatalf("request path = %q, want %q", requestPath, "/anthropic/v1/messages")
+	}
+}
+
 func TestClientStreamMessageParsesSSEEvents(t *testing.T) {
 	body := strings.Join([]string{
 		"event: message_start",
diff --git a/metadata/models/MiniMax-M2.7.json b/metadata/models/MiniMax-M2.7.json
new file mode 100644
index 00000000..361e239f
--- /dev/null
+++ b/metadata/models/MiniMax-M2.7.json
@@ -0,0 +1,10 @@
+{
+  "context_window": 204800,
+  "display_name": "MiniMax M2.7",
+  "description": "Text model with always-on reasoning.",
+  "supports_reasoning": true,
+  "supported_reasoning_levels": [
+    {"effort": "always_on", "description": "Always-on reasoning"}
+  ],
+  "input_modalities": ["text"]
+}
diff --git a/metadata/models/MiniMax-M3.json b/metadata/models/MiniMax-M3.json
new file mode 100644
index 00000000..6a6804a4
--- /dev/null
+++ b/metadata/models/MiniMax-M3.json
@@ -0,0 +1,11 @@
+{
+  "context_window": 1000000,
+  "display_name": "MiniMax M3",
+  "description": "Long-context multimodal model with adaptive reasoning.",
+  "supports_reasoning": true,
+  "supported_reasoning_levels": [
+    {"effort": "adaptive", "description": "Adaptive reasoning"},
+    {"effort": "disabled", "description": "Reasoning disabled"}
+  ],
+  "input_modalities": ["text", "image", "video"]
+}