From c900bbb67a3b875052dfa465c1559b63ee42f9cf Mon Sep 17 00:00:00 2001
From: Diego Rodrigues de Sa e Souza
 <8016841+diegosouzapw@users.noreply.github.com>
Date: Sat, 4 Jul 2026 00:37:10 -0300
Subject: [PATCH] fix(providers): strip thinking param for minimax-m2.7 on
 NVIDIA NIM (#6102)

Strip unsupported thinking param for minimax-m2.7 on NVIDIA NIM. Integrated into release/v3.8.44.
---
 CHANGELOG.md                                  |  2 +
 open-sse/translator/paramSupport.ts           |  4 ++
 .../nvidia-minimax-thinking-strip.test.ts     | 52 +++++++++++++++++++
 3 files changed, 58 insertions(+)
 create mode 100644 tests/unit/nvidia-minimax-thinking-strip.test.ts

diff --git a/CHANGELOG.md b/CHANGELOG.md
index 88471b5b6d..20339a8a0b 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -23,6 +23,8 @@
 
 ### 🔧 Bug Fixes
 
+- **fix(providers):** disable the unsupported `thinking` param for `minimax-m2.7` on NVIDIA NIM (the upstream rejects it). Regression guard: `tests/unit/nvidia-minimax-thinking-strip.test.ts`. (thanks @anki1kr)
+
 - **fix(mitm):** add an in-process guard so concurrent MITM server starts no longer race — a second start while one is already in flight is short-circuited instead of double-binding the listener. Regression guard: `tests/unit/mitm-start-guard.test.ts`. (thanks @anki1kr)
 
 - **translator (Responses → Chat Completions):** strip the Responses-API-only `truncation` field before forwarding a `/v1/responses` request to a non-OpenAI Chat Completions upstream. Strict upstreams (e.g. NVIDIA NIM) rejected it with HTTP 400 `Unsupported parameter(s): truncation`, breaking Codex-style clients routed to those providers. `client_metadata`, `background`, and `safety_identifier` were already stripped — `truncation` was the remaining gap. Regression guard: `tests/unit/responses-strip-truncation-2311.test.ts`. (thanks @TuanNguyen0708)
diff --git a/open-sse/translator/paramSupport.ts b/open-sse/translator/paramSupport.ts
index 142f5528bb..08e0e807a4 100644
--- a/open-sse/translator/paramSupport.ts
+++ b/open-sse/translator/paramSupport.ts
@@ -29,6 +29,10 @@ const STRIP_RULES: StripRule[] = [
       /claude/i.test(m) && !/claude.*(opus|sonnet).*4\.6/i.test(m),
     drop: ["thinking", "reasoning_effort"],
   },
+  // NVIDIA NIM minimaxai/minimax-m2.7: NVIDIA's OpenAI-compatible wrapper
+  // (format:"openai") does not accept the Claude-style `thinking` body field
+  // and returns 400 "Unsupported parameter(s): thinking". Upstream #2268.
+  { provider: "nvidia", match: /minimax-m2\.7/i, drop: ["thinking"] },
 ];
 
 function matches(rule: StripRule, model: string): boolean {
diff --git a/tests/unit/nvidia-minimax-thinking-strip.test.ts b/tests/unit/nvidia-minimax-thinking-strip.test.ts
new file mode 100644
index 0000000000..4c67616f39
--- /dev/null
+++ b/tests/unit/nvidia-minimax-thinking-strip.test.ts
@@ -0,0 +1,52 @@
+// Regression guard: NVIDIA NIM's OpenAI-compatible wrapper (format:"openai")
+// does not accept the Claude-style `thinking` body field for
+// minimaxai/minimax-m2.7 and returns 400 "Unsupported parameter(s): thinking".
+// Upstream #2268 / 9router#2323.
+//
+// Fix: paramSupport.ts STRIP_RULES drops `thinking` for provider "nvidia" +
+// model matching /minimax-m2\.7/i before the request reaches the executor.
+
+import { test } from "node:test";
+import assert from "node:assert/strict";
+import { stripUnsupportedParams } from "../../open-sse/translator/paramSupport.ts";
+
+test("stripUnsupportedParams: nvidia + minimaxai/minimax-m2.7 drops thinking", () => {
+  const body: Record<string, unknown> = {
+    model: "minimaxai/minimax-m2.7",
+    thinking: { type: "adaptive" },
+    max_tokens: 512,
+    temperature: 0.7,
+  };
+  stripUnsupportedParams("nvidia", "minimaxai/minimax-m2.7", body);
+  assert.equal(body.thinking, undefined, "thinking must be stripped for NVIDIA minimax-m2.7");
+  assert.equal(body.max_tokens, 512, "other params must survive");
+  assert.equal(body.temperature, 0.7, "other params must survive");
+  assert.equal(body.model, "minimaxai/minimax-m2.7", "model must not be touched");
+});
+
+test("stripUnsupportedParams: nvidia + other model KEEPS thinking (regression guard)", () => {
+  const body: Record<string, unknown> = { thinking: { type: "adaptive" } };
+  stripUnsupportedParams("nvidia", "some-other-model", body);
+  assert.deepEqual(
+    body.thinking,
+    { type: "adaptive" },
+    "unrelated NVIDIA-hosted models must not be affected"
+  );
+});
+
+test("stripUnsupportedParams: non-nvidia provider KEEPS thinking for minimax-m2.7", () => {
+  const body: Record<string, unknown> = { thinking: { type: "adaptive" } };
+  stripUnsupportedParams("minimax", "minimax-m2.7", body);
+  assert.deepEqual(
+    body.thinking,
+    { type: "adaptive" },
+    "direct MiniMax API still supports the native thinking field"
+  );
+});
+
+test("stripUnsupportedParams: nvidia + minimax-m2.7 without thinking present is a no-op", () => {
+  const body: Record<string, unknown> = { max_tokens: 100 };
+  stripUnsupportedParams("nvidia", "minimaxai/minimax-m2.7", body);
+  assert.equal(body.max_tokens, 100);
+  assert.equal("thinking" in body, false);
+});