diff --git a/packages/opencode/src/session/message-v2.ts b/packages/opencode/src/session/message-v2.ts
index a017ead1e631..1eed28418c07 100644
--- a/packages/opencode/src/session/message-v2.ts
+++ b/packages/opencode/src/session/message-v2.ts
@@ -854,13 +854,27 @@ export const toModelMessagesEffect = Effect.fnUntraced(function* (
         role: "assistant",
         parts: [],
       }
+      // Substitute a space for empty text between signed reasoning
+      // blocks to keep thinking block positions (and signatures) valid without
+      // triggering Anthropic's empty-content rejection or the AI SDK's filter.
+      // Signatures live under the provider-namespaced metadata key: anthropic
+      // (direct API), bedrock (AWS Bedrock), or vertex (GCP Vertex AI).
+      const hasSignedReasoning = msg.parts.some(
+        (p) =>
+          p.type === "reasoning" &&
+          ((p.metadata as any)?.anthropic?.signature != null ||
+            (p.metadata as any)?.bedrock?.signature != null ||
+            (p.metadata as any)?.vertex?.signature != null),
+      )
       for (const part of msg.parts) {
-        if (part.type === "text")
+        if (part.type === "text") {
+          const text = part.text === "" && hasSignedReasoning ? " " : part.text
           assistantMessage.parts.push({
             type: "text",
-            text: part.text,
+            text,
             ...(differentModel ? {} : { providerMetadata: part.metadata }),
           })
+        }
         if (part.type === "step-start")
           assistantMessage.parts.push({
             type: "step-start",
diff --git a/packages/opencode/test/session/message-v2.test.ts b/packages/opencode/test/session/message-v2.test.ts
index 89bae246a78c..b2590a7e32ed 100644
--- a/packages/opencode/test/session/message-v2.test.ts
+++ b/packages/opencode/test/session/message-v2.test.ts
@@ -1090,6 +1090,134 @@ describe("session.message-v2.toModelMessage", () => {
       },
     ])
   })
+
+  test("substitutes space for empty text between signed reasoning blocks", async () => {
+    // Reproduces the bug pattern: [reasoning(sig), text(""), reasoning(sig), text(full)]
+    const assistantID = "m-assistant"
+    const input: MessageV2.WithParts[] = [
+      {
+        info: assistantInfo(assistantID, "m-parent"),
+        parts: [
+          { ...basePart(assistantID, "p1"), type: "step-start" },
+          {
+            ...basePart(assistantID, "p2"),
+            type: "reasoning",
+            text: "thinking-one",
+            metadata: { anthropic: { signature: "sig1" } },
+          },
+          { ...basePart(assistantID, "p3"), type: "text", text: "" },
+          { ...basePart(assistantID, "p4"), type: "step-start" },
+          {
+            ...basePart(assistantID, "p5"),
+            type: "reasoning",
+            text: "thinking-two",
+            metadata: { anthropic: { signature: "sig2" } },
+          },
+          { ...basePart(assistantID, "p6"), type: "text", text: "the answer" },
+        ] as MessageV2.Part[],
+      },
+    ]
+
+    const result = await MessageV2.toModelMessages(input, model)
+
+    // step-start splits into two assistant messages; SDK's groupIntoBlocks merges them later
+    expect(result).toHaveLength(2)
+    expect((result[0].content as any[]).find((p) => p.type === "text").text).toBe(" ")
+    expect((result[1].content as any[]).find((p) => p.type === "text").text).toBe("the answer")
+  })
+
+  test("substitutes space for empty text when reasoning signature is under 'bedrock' namespace", async () => {
+    // AWS Bedrock hosts Anthropic Claude but stores signatures under metadata.bedrock
+    const assistantID = "m-assistant-bedrock"
+    const input: MessageV2.WithParts[] = [
+      {
+        info: assistantInfo(assistantID, "m-parent"),
+        parts: [
+          {
+            ...basePart(assistantID, "p1"),
+            type: "reasoning",
+            text: "thinking-bedrock",
+            metadata: { bedrock: { signature: "bedrock-sig" } },
+          },
+          { ...basePart(assistantID, "p2"), type: "text", text: "" },
+          { ...basePart(assistantID, "p3"), type: "text", text: "answer" },
+        ] as MessageV2.Part[],
+      },
+    ]
+
+    const result = await MessageV2.toModelMessages(input, model)
+
+    expect(result).toHaveLength(1)
+    const texts = (result[0].content as any[]).filter((p) => p.type === "text")
+    expect(texts.map((t) => t.text)).toStrictEqual([" ", "answer"])
+  })
+
+  test("substitutes space for empty text when reasoning signature is under 'vertex' namespace", async () => {
+    // GCP Vertex AI hosts Anthropic Claude but stores signatures under metadata.vertex
+    const assistantID = "m-assistant-vertex"
+    const input: MessageV2.WithParts[] = [
+      {
+        info: assistantInfo(assistantID, "m-parent"),
+        parts: [
+          {
+            ...basePart(assistantID, "p1"),
+            type: "reasoning",
+            text: "thinking-vertex",
+            metadata: { vertex: { signature: "vertex-sig" } },
+          },
+          { ...basePart(assistantID, "p2"), type: "text", text: "" },
+          { ...basePart(assistantID, "p3"), type: "text", text: "answer" },
+        ] as MessageV2.Part[],
+      },
+    ]
+
+    const result = await MessageV2.toModelMessages(input, model)
+
+    expect(result).toHaveLength(1)
+    const texts = (result[0].content as any[]).filter((p) => p.type === "text")
+    expect(texts.map((t) => t.text)).toStrictEqual([" ", "answer"])
+  })
+
+  test("leaves empty text alone when reasoning has no Anthropic signature", async () => {
+    // Non-Anthropic providers' reasoning doesn't position-validate, so empty text
+    // should be filtered normally rather than substituted.
+    const assistantID = "m-assistant-unsigned"
+    const input: MessageV2.WithParts[] = [
+      {
+        info: assistantInfo(assistantID, "m-parent"),
+        parts: [
+          { ...basePart(assistantID, "p1"), type: "reasoning", text: "thinking" },
+          { ...basePart(assistantID, "p2"), type: "text", text: "" },
+          { ...basePart(assistantID, "p3"), type: "text", text: "answer" },
+        ] as MessageV2.Part[],
+      },
+    ]
+
+    const result = await MessageV2.toModelMessages(input, model)
+
+    expect(result).toHaveLength(1)
+    const texts = (result[0].content as any[]).filter((p) => p.type === "text")
+    expect(texts.map((t) => t.text)).toStrictEqual(["", "answer"])
+  })
+
+  test("leaves empty text alone in assistant messages without reasoning", async () => {
+    const assistantID = "m-assistant-no-reasoning"
+    const input: MessageV2.WithParts[] = [
+      {
+        info: assistantInfo(assistantID, "m-parent"),
+        parts: [
+          { ...basePart(assistantID, "p1"), type: "text", text: "" },
+          { ...basePart(assistantID, "p2"), type: "text", text: "hello" },
+        ] as MessageV2.Part[],
+      },
+    ]
+
+    const result = await MessageV2.toModelMessages(input, model)
+
+    expect(result).toHaveLength(1)
+    const texts = (result[0].content as any[]).filter((p) => p.type === "text")
+    expect(texts.map((t) => t.text)).toStrictEqual(["", "hello"])
+  })
 })
 
 describe("session.message-v2.fromError", () => {