From bc73fa248181da3034ff2841f429231a71c04a84 Mon Sep 17 00:00:00 2001
From: OmniRoute Ops <ops@nomenak.dev>
Date: Tue, 12 May 2026 14:58:09 +0000
Subject: [PATCH] fix(reasoning-cache): include xiaomi-mimo in replay
 provider/model detection

MiMo (Xiaomi) enforces the same "echo reasoning_content on subsequent
turns" contract as DeepSeek and Kimi-thinking. Without replay, the
upstream returns 400:

  data:{"error":{"code":"400","message":"Param Incorrect",
   "param":"The reasoning_content in the thinking mode must be passed back to the API.","type":""}}

Repro: client sends a multi-turn /v1/messages body where the assistant
history has tool_use blocks but no thinking blocks (Capy and most BYOK
clients strip thinking on the wire). MiMo refuses without the
reasoning_content from the previous assistant turn.

The reasoning replay cache (issue #1628) already captures
reasoning_content from non-streaming responses with tool_calls and
re-injects it on the request side. But the gate
`requiresReasoningReplay(provider, model)` did not include MiMo:

  REASONING_REPLAY_PROVIDERS missed "xiaomi-mimo"
  REASONING_REPLAY_MODEL_PATTERNS had no /mimo/ entry

So the captured reasoning was discarded on the next turn instead of
replayed.

Fix:
  - Add "xiaomi-mimo" to REASONING_REPLAY_PROVIDERS
  - Add /^mimo[-.]?v\d/i to REASONING_REPLAY_MODEL_PATTERNS (defensive
    match if a wildcard route assigns a non-xiaomi-mimo provider ID to
    a mimo-* model alias)

Tests: 4 new cases (40/40 green) covering both provider-id and model-
pattern detection paths, including XIAOMI-MIMO uppercase normalization.
---
 open-sse/services/reasoningCache.ts |  8 ++++++++
 tests/unit/reasoning-cache.test.ts  | 14 ++++++++++++++
 2 files changed, 22 insertions(+)

diff --git a/open-sse/services/reasoningCache.ts b/open-sse/services/reasoningCache.ts
index 2b5f7788e37..95da644d27a 100644
--- a/open-sse/services/reasoningCache.ts
+++ b/open-sse/services/reasoningCache.ts
@@ -34,6 +34,11 @@ const REASONING_REPLAY_PROVIDERS = new Set([
   "sambanova",
   "fireworks",
   "together",
+  // Xiaomi MiMo enforces the same "pass back reasoning_content on subsequent
+  // turns" contract as DeepSeek/Kimi-thinking. Without replay the upstream
+  // 400s with "Param Incorrect: The reasoning_content in the thinking mode
+  // must be passed back to the API."
+  "xiaomi-mimo",
 ]);
 
 const REASONING_REPLAY_MODEL_PATTERNS = [
@@ -44,6 +49,9 @@ const REASONING_REPLAY_MODEL_PATTERNS = [
   /qwq/i,
   /qwen.*think/i,
   /glm.*think/i,
+  // MiMo (Xiaomi) thinking models — defensive match if a wildcard route
+  // assigns a non-`xiaomi-mimo` provider ID to a mimo-* model alias.
+  /^mimo[-.]?v\d/i,
 ];
 
 /**
diff --git a/tests/unit/reasoning-cache.test.ts b/tests/unit/reasoning-cache.test.ts
index b2dcf6d496a..7fa835239f9 100644
--- a/tests/unit/reasoning-cache.test.ts
+++ b/tests/unit/reasoning-cache.test.ts
@@ -346,6 +346,20 @@ describe("Reasoning Replay Cache — Provider Detection", () => {
     assert.equal(requiresReasoningReplay("glm", "glm-5-thinking"), true);
   });
 
+  it("should detect xiaomi-mimo provider", () => {
+    // MiMo enforces reasoning_content echo on subsequent turns; without
+    // replay the upstream returns 400 "Param Incorrect: The reasoning_content
+    // in the thinking mode must be passed back to the API."
+    assert.equal(requiresReasoningReplay("xiaomi-mimo", "mimo-v2.5-pro"), true);
+    assert.equal(requiresReasoningReplay("XIAOMI-MIMO", "mimo-v2.5"), true);
+  });
+
+  it("should detect mimo-v* model pattern under any provider id", () => {
+    assert.equal(requiresReasoningReplay("unknown-provider", "mimo-v2.5-pro"), true);
+    assert.equal(requiresReasoningReplay("unknown-provider", "mimo-v3"), true);
+    assert.equal(requiresReasoningReplay("unknown-provider", "MimoV2.5-pro"), true);
+  });
+
   it("should NOT detect a generic openai model", () => {
     assert.equal(requiresReasoningReplay("openai", "gpt-4o"), false);
   });