diff --git a/changelog.d/fixes/10940-opencode-limit-output.md b/changelog.d/fixes/10940-opencode-limit-output.md
new file mode 100644
index 0000000000..9af54a2046
--- /dev/null
+++ b/changelog.d/fixes/10940-opencode-limit-output.md
@@ -0,0 +1 @@
+- fix(cli): always emit limit.output in generated OpenCode config so schema validation passes for metadata-less models (#10940)
diff --git a/src/lib/cli-helper/config-generator/opencode.ts b/src/lib/cli-helper/config-generator/opencode.ts
index c2915e5eb7..e3d3f0e950 100644
--- a/src/lib/cli-helper/config-generator/opencode.ts
+++ b/src/lib/cli-helper/config-generator/opencode.ts
@@ -243,9 +243,11 @@ function resolveContextLength(entry: CatalogModelEntry): number | undefined {
  *   1. Existing manual override in the user's opencode.json (`limit.context`).
  *   2. Catalog `context_length` / `max_context_window_tokens`.
  *
- * If neither is available, the entry is returned WITHOUT a `limit` block so
- * the caller can decide whether to skip the model entirely or surface a
- * warning. We never fabricate a default context window.
+ * If neither is available, `limit.context` is simply omitted and OpenCode's
+ * own heuristics apply — we never fabricate a default context window. The
+ * entry ALWAYS carries a `limit` block, though: `limit.output` is a
+ * required field in OpenCode's v1 provider schema, so it is always emitted
+ * (falling back to 8K when nothing else is known) — see #10940.
  */
 function buildModelEntry(
   id: string,
@@ -301,25 +303,23 @@ function buildModelEntry(
   const output =
     typeof userOutput === "number" && userOutput > 0 ? userOutput : (catalogOutput ?? 8_192);
 
-  // Emit `limit` only if we have at least one of context/output. We never
-  // emit a half-baked limit block with only an `output` (would be misleading).
-  if (
-    typeof context === "number" ||
-    typeof userOutput === "number" ||
-    typeof catalogOutput === "number"
-  ) {
-    const limit: { context?: number; input?: number; output?: number } = {};
-    if (typeof context === "number") limit.context = context;
-    limit.output = output;
-    const userInput = existing?.limit?.input;
-    if (typeof userInput === "number" && userInput > 0) {
-      limit.input = userInput;
-    } else if (catalog) {
-      const maxInput = catalog.max_input_tokens;
-      if (typeof maxInput === "number" && maxInput > 0) limit.input = maxInput;
-    }
-    entry.limit = limit;
+  // `limit.output` is REQUIRED by OpenCode's v1 provider schema regardless of
+  // whether the catalog (or the user's existing config) knows the model's
+  // context window — a model with no catalog metadata at all must still get
+  // a `limit` block, or OpenCode rejects the whole config with "Missing key
+  // provider.omniroute.models.{model}.limit.output" (#10940). `output` above
+  // already resolves to a safe fallback (8K) when nothing else is known, so
+  // we always emit it; `context`/`input` are added only when actually known.
+  const limit: { context?: number; input?: number; output?: number } = { output };
+  if (typeof context === "number") limit.context = context;
+  const userInput = existing?.limit?.input;
+  if (typeof userInput === "number" && userInput > 0) {
+    limit.input = userInput;
+  } else if (catalog) {
+    const maxInput = catalog.max_input_tokens;
+    if (typeof maxInput === "number" && maxInput > 0) limit.input = maxInput;
   }
+  entry.limit = limit;
 
   return entry;
 }
diff --git a/tests/unit/cli-helper/config-generator.test.ts b/tests/unit/cli-helper/config-generator.test.ts
index fde409b940..1b2e85732f 100644
--- a/tests/unit/cli-helper/config-generator.test.ts
+++ b/tests/unit/cli-helper/config-generator.test.ts
@@ -603,12 +603,15 @@ describe("config-generator", () => {
           input: 100000,
           output: 32768,
         });
-        assert.strictEqual(models["no-metadata"].limit, undefined);
+        // #10940: `limit.output` is REQUIRED by OpenCode's v1 provider schema,
+        // so even a model with zero catalog metadata still gets a `limit`
+        // block carrying the fallback output value; `context`/`input` stay
+        // omitted since neither the catalog nor the user knows them.
+        assert.deepStrictEqual(models["no-metadata"].limit, { output: 8192 });
 
         for (const model of Object.values(models) as Array<{ limit?: { output?: number } }>) {
           assert.ok(
-            model.limit === undefined ||
-              (typeof model.limit.output === "number" && model.limit.output > 0),
+            typeof model.limit?.output === "number" && model.limit.output > 0,
             "every emitted limit must contain a positive output"
           );
         }
diff --git a/tests/unit/opencode-limit-output-10940.test.ts b/tests/unit/opencode-limit-output-10940.test.ts
new file mode 100644
index 0000000000..653002ebe8
--- /dev/null
+++ b/tests/unit/opencode-limit-output-10940.test.ts
@@ -0,0 +1,85 @@
+import { describe, it } from "node:test";
+import assert from "node:assert";
+
+/**
+ * Regression guard for #10940: OpenCode rejects a generated config with
+ * "Missing key provider.omniroute.models.{model}.limit.output" whenever a
+ * model has no catalog metadata (no `context_length`, no
+ * `max_output_tokens`) and no existing user override. `limit.output` is a
+ * REQUIRED field in OpenCode's v1 provider schema, so it must always be
+ * emitted — even when nothing is known about the model.
+ */
+describe("opencode config generator — limit.output always emitted (#10940)", () => {
+  function makeCatalogResponse(models: unknown[]): unknown {
+    return { object: "list", data: models };
+  }
+
+  function stubFetchOnce(body: unknown, status = 200) {
+    const original = globalThis.fetch;
+    // @ts-ignore — globalThis.fetch signature is compatible for our purposes
+    globalThis.fetch = (async () => {
+      return new Response(JSON.stringify(body), {
+        status,
+        headers: { "content-type": "application/json" },
+      });
+    }) as typeof fetch;
+    return {
+      restore: () => {
+        globalThis.fetch = original;
+      },
+    };
+  }
+
+  it("RED-proving case: a model with NO context_length and NO max_output_tokens still gets a numeric limit.output", async () => {
+    // This model has no metadata whatsoever beyond its id — exactly the
+    // shape that used to leave `entry.limit` undefined entirely (issue #10940).
+    const stub = stubFetchOnce(
+      makeCatalogResponse([{ id: "metadataless-model", owned_by: "someProvider" }])
+    );
+    try {
+      const { generateOpencodeConfig } = await import(
+        "../../src/lib/cli-helper/config-generator/opencode.ts"
+      );
+      const out = await generateOpencodeConfig({
+        baseUrl: "http://localhost:20128",
+        apiKey: "sk-test",
+      });
+      const cfg = JSON.parse(out);
+      const entry = cfg.provider.omniroute.models["metadataless-model"];
+      assert.ok(entry, "model entry must exist in the generated config");
+      assert.ok(entry.limit, "entry.limit must be present even without catalog metadata");
+      assert.strictEqual(
+        typeof entry.limit.output,
+        "number",
+        `entry.limit.output must be a number, got ${JSON.stringify(entry.limit?.output)}`
+      );
+      assert.ok(entry.limit.output > 0, "entry.limit.output must be a positive number");
+      // context stays unknown — we must NOT fabricate it.
+      assert.strictEqual(entry.limit.context, undefined);
+    } finally {
+      stub.restore();
+    }
+  });
+
+  it("honors the catalog's max_output_tokens when present", async () => {
+    const stub = stubFetchOnce(
+      makeCatalogResponse([
+        { id: "has-output-meta", owned_by: "someProvider", max_output_tokens: 4096 },
+      ])
+    );
+    try {
+      const { generateOpencodeConfig } = await import(
+        "../../src/lib/cli-helper/config-generator/opencode.ts"
+      );
+      const out = await generateOpencodeConfig({
+        baseUrl: "http://localhost:20128",
+        apiKey: "sk-test",
+      });
+      const cfg = JSON.parse(out);
+      const entry = cfg.provider.omniroute.models["has-output-meta"];
+      assert.strictEqual(entry.limit.output, 4096);
+    } finally {
+      stub.restore();
+    }
+  });
+});