diff --git a/open-sse/config/providers/registry/theoldllm/index.ts b/open-sse/config/providers/registry/theoldllm/index.ts
index 947bfa67f3..1b1c7691cb 100644
--- a/open-sse/config/providers/registry/theoldllm/index.ts
+++ b/open-sse/config/providers/registry/theoldllm/index.ts
@@ -10,15 +10,16 @@ export const theoldllmProvider: RegistryEntry = {
   baseUrls: ["https://theoldllm.vercel.app/api/chatgpt"],
   authType: "none",
   authHeader: "none",
+  defaultContextLength: 200000,
   models: [
-    { id: "GPT_5_4", name: "GPT-5.4 (The Old LLM 🆓)" },
+    { id: "GPT_5_4", name: "GPT-5.4 (The Old LLM 🆓)", contextLength: 400000 },
     { id: "GPT_4o", name: "GPT-4o (The Old LLM 🆓)" },
-    { id: "claude_opus_4", name: "Claude Opus 4 (The Old LLM 🆓)" },
-    { id: "claude_sonnet_4", name: "Claude Sonnet 4 (The Old LLM 🆓)" },
-    { id: "claude_haiku_3_5", name: "Claude Haiku 3.5 (The Old LLM 🆓)" },
-    { id: "deepseek_v4", name: "DeepSeek V4 (The Old LLM 🆓)" },
-    { id: "gemini_3_flash", name: "Gemini 3 Flash (The Old LLM 🆓)" },
-    { id: "gemini_3_pro", name: "Gemini 3 Pro (The Old LLM 🆓)" },
+    { id: "claude_opus_4", name: "Claude Opus 4 (The Old LLM 🆓)", contextLength: 200000 },
+    { id: "claude_sonnet_4", name: "Claude Sonnet 4 (The Old LLM 🆓)", contextLength: 200000 },
+    { id: "claude_haiku_3_5", name: "Claude Haiku 3.5 (The Old LLM 🆓)", contextLength: 200000 },
+    { id: "deepseek_v4", name: "DeepSeek V4 (The Old LLM 🆓)", contextLength: 200000 },
+    { id: "gemini_3_flash", name: "Gemini 3 Flash (The Old LLM 🆓)", contextLength: 1000000 },
+    { id: "gemini_3_pro", name: "Gemini 3 Pro (The Old LLM 🆓)", contextLength: 1000000 },
   ],
   passthroughModels: true,
 };
diff --git a/tests/unit/theoldllm-context-length-4184.test.ts b/tests/unit/theoldllm-context-length-4184.test.ts
new file mode 100644
index 0000000000..385b635ac0
--- /dev/null
+++ b/tests/unit/theoldllm-context-length-4184.test.ts
@@ -0,0 +1,57 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+// Regression guard for #4184.
+//
+// The theoldllm provider (free OpenAI-compatible upstream) listed its models
+// with NO contextLength, so getResolvedModelCapabilities resolved their context
+// window to `null` and the dashboard/catalog reported no usable window. #4184
+// adds an entry-level `defaultContextLength` plus per-model `contextLength`
+// overrides reflecting each upstream model's real window. This test asserts both
+// the registry data (source of truth) and the resolved context window for the
+// models that carry an explicit override — the latter would resolve to `null`
+// on the pre-#4184 registry, so it fails without the fix.
+const { getRegistryEntry } = await import("../../open-sse/config/providerRegistry.ts");
+const { getResolvedModelCapabilities } = await import("../../src/lib/modelCapabilities.ts");
+
+function model(id: string) {
+  const entry = getRegistryEntry("theoldllm");
+  assert.ok(entry, "theoldllm registry entry must exist");
+  return (entry.models ?? []).find((m) => m.id === id);
+}
+
+test("#4184 theoldllm entry declares a 200000 defaultContextLength", () => {
+  const entry = getRegistryEntry("theoldllm");
+  assert.ok(entry, "theoldllm registry entry must exist");
+  assert.equal(entry.defaultContextLength, 200000);
+});
+
+test("#4184 per-model contextLength overrides match each upstream window", () => {
+  assert.equal(model("GPT_5_4")?.contextLength, 400000, "GPT-5.4 window is 400K");
+  assert.equal(model("gemini_3_flash")?.contextLength, 1000000, "Gemini 3 Flash window is 1M");
+  assert.equal(model("gemini_3_pro")?.contextLength, 1000000, "Gemini 3 Pro window is 1M");
+  for (const id of ["claude_opus_4", "claude_sonnet_4", "claude_haiku_3_5", "deepseek_v4"]) {
+    assert.equal(model(id)?.contextLength, 200000, `${id} window is 200K`);
+  }
+});
+
+test("#4184 GPT_4o carries no explicit contextLength (relies on defaultContextLength)", () => {
+  // Intentionally left to the entry default — documents the fallback contract so a
+  // later edit that removes defaultContextLength is caught by the assertion above.
+  assert.equal(model("GPT_4o")?.contextLength, undefined);
+});
+
+test("#4184 resolved context window reflects the override (null before the fix)", () => {
+  assert.equal(
+    getResolvedModelCapabilities({ provider: "theoldllm", model: "GPT_5_4" }).contextWindow,
+    400000
+  );
+  assert.equal(
+    getResolvedModelCapabilities({ provider: "theoldllm", model: "gemini_3_pro" }).contextWindow,
+    1000000
+  );
+  assert.equal(
+    getResolvedModelCapabilities({ provider: "theoldllm", model: "claude_opus_4" }).contextWindow,
+    200000
+  );
+});