diff --git a/extensions/deepinfra/index.ts b/extensions/deepinfra/index.ts
index e0f19efe415..b4b7c5f4d86 100644
--- a/extensions/deepinfra/index.ts
+++ b/extensions/deepinfra/index.ts
@@ -1,7 +1,11 @@
 import { definePluginEntry } from "openclaw/plugin-sdk/core";
 import { createProviderApiKeyAuthMethod } from "openclaw/plugin-sdk/provider-auth";
 import { buildSingleProviderApiKeyCatalog } from "openclaw/plugin-sdk/provider-catalog";
-import { createDeepInfraSystemCacheWrapper } from "openclaw/plugin-sdk/provider-stream";
+import {
+  createDeepInfraSystemCacheWrapper,
+  createDeepInfraWrapper,
+  isProxyReasoningUnsupported,
+} from "openclaw/plugin-sdk/provider-stream";
 import { applyDeepInfraConfig, DEEPINFRA_DEFAULT_MODEL_REF } from "./onboard.js";
 import { buildDeepInfraProviderWithDiscovery } from "./provider-catalog.js";
 
@@ -66,7 +70,14 @@ export default definePluginEntry({
         geminiThoughtSignatureModelHints: ["gemini"],
         dropThinkingBlockModelHints: ["claude"],
       },
-      wrapStreamFn: (ctx) => createDeepInfraSystemCacheWrapper(ctx.streamFn),
+      wrapStreamFn: (ctx) => {
+        const thinkingLevel = isProxyReasoningUnsupported(ctx.modelId)
+          ? undefined
+          : ctx.thinkingLevel;
+        let streamFn = createDeepInfraWrapper(ctx.streamFn, thinkingLevel);
+        streamFn = createDeepInfraSystemCacheWrapper(streamFn);
+        return streamFn;
+      },
       isCacheTtlEligible: (ctx) => isDeepInfraCacheTtlModel(ctx.modelId),
     });
   },
diff --git a/src/agents/pi-embedded-runner/proxy-stream-wrappers.test.ts b/src/agents/pi-embedded-runner/proxy-stream-wrappers.test.ts
index 487d90582ef..54a39a53aee 100644
--- a/src/agents/pi-embedded-runner/proxy-stream-wrappers.test.ts
+++ b/src/agents/pi-embedded-runner/proxy-stream-wrappers.test.ts
@@ -2,7 +2,7 @@ import type { StreamFn } from "@mariozechner/pi-agent-core";
 import type { Context, Model } from "@mariozechner/pi-ai";
 import { createAssistantMessageEventStream } from "@mariozechner/pi-ai";
 import { describe, expect, it } from "vitest";
-import { createOpenRouterWrapper } from "./proxy-stream-wrappers.js";
+import { createDeepInfraWrapper, createOpenRouterWrapper } from "./proxy-stream-wrappers.js";
 
 describe("proxy stream wrappers", () => {
   it("adds OpenRouter attribution headers to stream options", () => {
@@ -35,4 +35,67 @@ describe("proxy stream wrappers", () => {
       },
     ]);
   });
+
+  describe("createDeepInfraWrapper", () => {
+    function capturePayloads() {
+      const payloads: unknown[] = [];
+      const baseStreamFn: StreamFn = (_model, _context, options) => {
+        const payload = { model: "test" };
+        options?.onPayload?.(payload, _model);
+        payloads.push(structuredClone(payload));
+        return createAssistantMessageEventStream();
+      };
+      return { baseStreamFn, payloads };
+    }
+
+    const model = {
+      api: "openai-completions",
+      provider: "deepinfra",
+      id: "moonshotai/Kimi-K2.5",
+    } as Model<"openai-completions">;
+    const context: Context = { messages: [] };
+
+    it("injects reasoning effort when thinkingLevel is set", () => {
+      const { baseStreamFn, payloads } = capturePayloads();
+      const wrapped = createDeepInfraWrapper(baseStreamFn, "high");
+      void wrapped(model, context, {});
+
+      expect(payloads[0]).toEqual({
+        model: "test",
+        reasoning: { effort: "high" },
+      });
+    });
+
+    it("maps 'off' to no reasoning field", () => {
+      const { baseStreamFn, payloads } = capturePayloads();
+      const wrapped = createDeepInfraWrapper(baseStreamFn, "off");
+      void wrapped(model, context, {});
+
+      expect(payloads[0]).toEqual({ model: "test" });
+    });
+
+    it("does not inject reasoning when thinkingLevel is undefined", () => {
+      const { baseStreamFn, payloads } = capturePayloads();
+      const wrapped = createDeepInfraWrapper(baseStreamFn, undefined);
+      void wrapped(model, context, {});
+
+      expect(payloads[0]).toEqual({ model: "test" });
+    });
+
+    it("preserves existing onPayload callback", () => {
+      const { baseStreamFn } = capturePayloads();
+      const wrapped = createDeepInfraWrapper(baseStreamFn, "low");
+      const seen: unknown[] = [];
+      void wrapped(model, context, {
+        onPayload: (payload) => {
+          seen.push(structuredClone(payload));
+        },
+      });
+
+      expect(seen[0]).toEqual({
+        model: "test",
+        reasoning: { effort: "low" },
+      });
+    });
+  });
 });
diff --git a/src/agents/pi-embedded-runner/proxy-stream-wrappers.ts b/src/agents/pi-embedded-runner/proxy-stream-wrappers.ts
index 50568ba6efd..70ea42bec41 100644
--- a/src/agents/pi-embedded-runner/proxy-stream-wrappers.ts
+++ b/src/agents/pi-embedded-runner/proxy-stream-wrappers.ts
@@ -135,6 +135,23 @@ export function isProxyReasoningUnsupported(modelId: string): boolean {
   return modelId.toLowerCase().startsWith("x-ai/");
 }
 
+export function createDeepInfraWrapper(
+  baseStreamFn: StreamFn | undefined,
+  thinkingLevel?: ThinkLevel,
+): StreamFn {
+  const underlying = baseStreamFn ?? streamSimple;
+  return (model, context, options) => {
+    const onPayload = options?.onPayload;
+    return underlying(model, context, {
+      ...options,
+      onPayload: (payload) => {
+        normalizeProxyReasoningPayload(payload, thinkingLevel);
+        return onPayload?.(payload, model);
+      },
+    });
+  };
+}
+
 export function createKilocodeWrapper(
   baseStreamFn: StreamFn | undefined,
   thinkingLevel?: ThinkLevel,
diff --git a/src/plugin-sdk/provider-stream.ts b/src/plugin-sdk/provider-stream.ts
index 68df86e5116..74f9484e8e9 100644
--- a/src/plugin-sdk/provider-stream.ts
+++ b/src/plugin-sdk/provider-stream.ts
@@ -10,6 +10,7 @@ export {
 } from "../agents/pi-embedded-runner/google-stream-wrappers.js";
 export {
   createDeepInfraSystemCacheWrapper,
+  createDeepInfraWrapper,
   createKilocodeWrapper,
   createOpenRouterSystemCacheWrapper,
   createOpenRouterWrapper,