diff --git a/packages/pi-ai/scripts/generate-models.ts b/packages/pi-ai/scripts/generate-models.ts
index a36dacfef..a35300b85 100644
--- a/packages/pi-ai/scripts/generate-models.ts
+++ b/packages/pi-ai/scripts/generate-models.ts
@@ -268,6 +268,15 @@ function applyThinkingLevelMetadata(model: Model<any>): void {
 	) {
 		mergeAnthropicMessagesCompat(model, { forceAdaptiveThinking: true });
 	}
+	if (
+		(model.provider === "minimax" || model.provider === "minimax-cn") &&
+		model.api === "anthropic-messages" &&
+		model.id === "MiniMax-M3"
+	) {
+		// M3 exposes adaptive and disabled thinking modes over the Anthropic-compatible
+		// endpoint, so force adaptive thinking instead of the budget-based default.
+		mergeAnthropicMessagesCompat(model, { forceAdaptiveThinking: true });
+	}
 	if (model.api === "openai-completions" && model.id.includes("deepseek-v4")) {
 		mergeThinkingLevelMap(model, DEEPSEEK_V4_THINKING_LEVEL_MAP);
 	}
@@ -1786,12 +1795,16 @@ async function generateModels() {
 		}
 	}
 
-	const minimaxDirectModelOverrides = new Map([
+	const minimaxDirectModelOverrides = new Map<
+		string,
+		{ contextWindow: number; maxTokens: number; input?: ("text" | "image" | "video")[] }
+	>([
 		["MiniMax-M2.7", { contextWindow: 204800, maxTokens: 131072 }],
 		["MiniMax-M2.7-highspeed", { contextWindow: 204800, maxTokens: 131072 }],
 		// MiniMax's API overview advertises a 1M context window for M3; models.dev
-		// currently reports the documented guaranteed 512K floor.
-		["MiniMax-M3", { contextWindow: 1000000, maxTokens: 131072 }],
+		// currently reports the documented guaranteed 512K floor. M3 also accepts
+		// video inputs alongside text and image, which models.dev does not report.
+		["MiniMax-M3", { contextWindow: 1000000, maxTokens: 131072, input: ["text", "image", "video"] }],
 	]);
 	const minimaxDirectSupportedIds = new Set(minimaxDirectModelOverrides.keys());
 
@@ -1804,6 +1817,9 @@ async function generateModels() {
 			if (override) {
 				candidate.contextWindow = override.contextWindow;
 				candidate.maxTokens = override.maxTokens;
+				if (override.input) {
+					candidate.input = override.input;
+				}
 			}
 		}
 	}
diff --git a/packages/pi-ai/src/model-catalog.ts b/packages/pi-ai/src/model-catalog.ts
index d998f9a54..c34986491 100644
--- a/packages/pi-ai/src/model-catalog.ts
+++ b/packages/pi-ai/src/model-catalog.ts
@@ -111,7 +111,7 @@ const ModelCatalogEntrySchema = Type.Object({
 	baseUrl: Type.String(),
 	reasoning: Type.Boolean(),
 	thinkingLevelMap: Type.Optional(ThinkingLevelMapSchema),
-	input: Type.Array(Type.Union([Type.Literal("text"), Type.Literal("image")]), { minItems: 1 }),
+	input: Type.Array(Type.Union([Type.Literal("text"), Type.Literal("image"), Type.Literal("video")]), { minItems: 1 }),
 	cost: Type.Object({
 		input: Type.Number(),
 		output: Type.Number(),
diff --git a/packages/pi-ai/src/models.generated.json b/packages/pi-ai/src/models.generated.json
index ffc2230fe..5df2a079f 100644
--- a/packages/pi-ai/src/models.generated.json
+++ b/packages/pi-ai/src/models.generated.json
@@ -7694,10 +7694,14 @@
       "api": "anthropic-messages",
       "provider": "minimax",
       "baseUrl": "https://api.minimax.io/anthropic",
+      "compat": {
+        "forceAdaptiveThinking": true
+      },
       "reasoning": true,
       "input": [
         "text",
-        "image"
+        "image",
+        "video"
       ],
       "cost": {
         "input": 0.6,
@@ -7754,10 +7758,14 @@
       "api": "anthropic-messages",
       "provider": "minimax-cn",
       "baseUrl": "https://api.minimaxi.com/anthropic",
+      "compat": {
+        "forceAdaptiveThinking": true
+      },
       "reasoning": true,
       "input": [
         "text",
-        "image"
+        "image",
+        "video"
       ],
       "cost": {
         "input": 0.6,
diff --git a/packages/pi-ai/src/models.generated.ts b/packages/pi-ai/src/models.generated.ts
index 0ded61628..a5db0f87d 100644
--- a/packages/pi-ai/src/models.generated.ts
+++ b/packages/pi-ai/src/models.generated.ts
@@ -6132,8 +6132,9 @@ export const MODELS = {
 			api: "anthropic-messages",
 			provider: "minimax",
 			baseUrl: "https://api.minimax.io/anthropic",
+			compat: {"forceAdaptiveThinking":true},
 			reasoning: true,
-			input: ["text", "image"],
+			input: ["text", "image", "video"],
 			cost: {
 				input: 0.6,
 				output: 2.4,
@@ -6185,8 +6186,9 @@ export const MODELS = {
 			api: "anthropic-messages",
 			provider: "minimax-cn",
 			baseUrl: "https://api.minimaxi.com/anthropic",
+			compat: {"forceAdaptiveThinking":true},
 			reasoning: true,
-			input: ["text", "image"],
+			input: ["text", "image", "video"],
 			cost: {
 				input: 0.6,
 				output: 2.4,
diff --git a/packages/pi-ai/src/types.ts b/packages/pi-ai/src/types.ts
index d839b5724..864a9e6b1 100644
--- a/packages/pi-ai/src/types.ts
+++ b/packages/pi-ai/src/types.ts
@@ -609,7 +609,7 @@ export interface Model<TApi extends Api> {
 	 * Missing keys use provider defaults. null marks a level as unsupported.
 	 */
 	thinkingLevelMap?: ThinkingLevelMap;
-	input: ("text" | "image")[];
+	input: ("text" | "image" | "video")[];
 	cost: {
 		input: number; // $/million tokens
 		output: number; // $/million tokens
diff --git a/packages/pi-ai/test/generated-models.test.ts b/packages/pi-ai/test/generated-models.test.ts
index 763eb2d95..6b584eb91 100644
--- a/packages/pi-ai/test/generated-models.test.ts
+++ b/packages/pi-ai/test/generated-models.test.ts
@@ -180,7 +180,8 @@ describe("models.generated.ts", () => {
 				provider,
 				baseUrl,
 				reasoning: true,
-				input: ["text", "image"],
+				input: ["text", "image", "video"],
+				compat: { forceAdaptiveThinking: true },
 				cost: {
 					input: 0.6,
 					output: 2.4,
diff --git a/packages/pi-coding-agent/src/core/model-registry.ts b/packages/pi-coding-agent/src/core/model-registry.ts
index 391e611e2..0594f9bc2 100644
--- a/packages/pi-coding-agent/src/core/model-registry.ts
+++ b/packages/pi-coding-agent/src/core/model-registry.ts
@@ -1167,7 +1167,7 @@ export class ModelRegistry {
 	private getDiscoveryProviderDefaults(provider: string): {
 		api: Api;
 		baseUrl: string;
-		input: ("text" | "image")[];
+		input: Model<Api>["input"];
 		contextWindow: number;
 		maxTokens: number;
 	} {