feat(nvidia): add new models and capabilities for NVIDIA provider

- Updated capabilities for NVIDIA models to enforce OpenAI-compatible reasoning formats.
- Added new models: MiniMax M3, GLM 5.2, DeepSeek V4 Pro, DeepSeek V4 Flash, Kimi K2.6, and Nemotron 3 Ultra to the NVIDIA registry.

This enhances the provider's functionality and aligns with OpenAI standards.

open-sse/providers/capabilities.js

@@ -102,6 +102,15 @@ export const MODEL_CAPABILITIES = {
  * Provider-specific capability overrides. Keyed by provider alias/id.
  */
 export const PROVIDER_CAPABILITIES = {
+  // NVIDIA NIM is OpenAI-compatible → rejects MiniMax/GLM native `thinking` field.
+  // Force openai reasoning_effort format for its reasoning models. #issue
+  "nvidia": {
+    "minimaxai/minimax-m2.7": { reasoning: true, thinkingFormat: "openai", thinkingCanDisable: false, contextWindow: 200000, maxOutput: 131072 },
+    "minimaxai/minimax-m3": { vision: true, reasoning: true, thinkingFormat: "openai", thinkingCanDisable: false, contextWindow: 512000, maxOutput: 131072 },
+    "z-ai/glm-5.2": { reasoning: true, thinkingFormat: "openai", contextWindow: 200000, maxOutput: 128000 },
+    "deepseek-ai/deepseek-v4-pro": { reasoning: true, thinkingFormat: "openai", contextWindow: 1000000, maxOutput: 65536 },
+    "deepseek-ai/deepseek-v4-flash": { reasoning: true, thinkingFormat: "openai", contextWindow: 1000000, maxOutput: 65536 },
+  },
   // CodeBuddy.cn — authoritative per-model metadata from the gateway's model
   // config (contextWindow=maxInputTokens, maxOutput=maxOutputTokens, vision=
   // supportsImages). Every model reasons via OpenAI-style reasoning_effort

open-sse/providers/registry/nvidia.js

@@ -20,8 +20,13 @@ export default {
     validateUrl: "https://integrate.api.nvidia.com/v1/models",
   },
   models: [
-    { id: "minimaxai/minimax-m2.7", name: "Minimax M2.7" },
-    { id: "z-ai/glm4.7", name: "GLM 4.7" },
+    { id: "minimaxai/minimax-m2.7", name: "MiniMax M2.7" },
+    { id: "minimaxai/minimax-m3", name: "MiniMax M3" },
+    { id: "z-ai/glm-5.2", name: "GLM 5.2" },
+    { id: "deepseek-ai/deepseek-v4-pro", name: "DeepSeek V4 Pro" },
+    { id: "deepseek-ai/deepseek-v4-flash", name: "DeepSeek V4 Flash" },
+    { id: "moonshotai/kimi-k2.6", name: "Kimi K2.6" },
+    { id: "nvidia/nemotron-3-ultra-550b-a55b", name: "Nemotron 3 Ultra" },
     { id: "nvidia/nv-embedqa-e5-v5", name: "NV EmbedQA E5 v5", kind: "embedding" },
     { id: "nvidia/parakeet-ctc-1.1b-asr", name: "Parakeet CTC 1.1B", params: ["language"], kind: "stt" },
     { id: "fastpitch", name: "FastPitch", kind: "tts" },

tests/translator/real/nvidia-thinking.e2e.test.js

@@ -0,0 +1,58 @@
+// E2E: hit live local proxy → verify nvidia MiniMax M2.7 doesn't 400 on
+// unsupported "thinking" param (nvidia NIM is OpenAI-compatible).
+// Requires dev server running on NV_E2E_PORT + an active router API key in DB.
+// RUN_E2E=1 npx vitest run --config tests/vitest.config.js tests/translator/real/nvidia-thinking.e2e.test.js
+import { describe, it, expect, beforeAll } from "vitest";
+import { getApiKeys } from "../../../src/lib/db/repos/apiKeysRepo.js";
+
+const PORT = process.env.NV_E2E_PORT || "20127";
+const BASE = `http://localhost:${PORT}`;
+const MODELS = [
+  "nvidia/minimaxai/minimax-m2.7",
+  "nvidia/minimaxai/minimax-m3",
+  "nvidia/z-ai/glm-5.2",
+  "nvidia/deepseek-ai/deepseek-v4-pro",
+  "nvidia/deepseek-ai/deepseek-v4-flash",
+  "nvidia/moonshotai/kimi-k2.6",
+  "nvidia/nvidia/nemotron-3-ultra-550b-a55b",
+];
+const RUN = process.env.RUN_E2E === "1";
+const maybe = RUN ? describe : describe.skip;
+
+async function drain(res) {
+  const reader = res.body.getReader();
+  const decoder = new TextDecoder();
+  let out = "";
+  while (true) {
+    const { done, value } = await reader.read();
+    if (done) break;
+    out += decoder.decode(value, { stream: true });
+  }
+  return out;
+}
+
+maybe("nvidia thinking e2e", () => {
+  let apiKey = "";
+  beforeAll(async () => {
+    const keys = await getApiKeys();
+    apiKey = keys.find((k) => k.isActive)?.key || process.env.NV_E2E_KEY || "";
+  });
+
+  it.each(MODELS)("%s with reasoning_effort -> no 'thinking' 400", async (model) => {
+    if (!apiKey) return expect(true).toBe(true);
+    const res = await fetch(`${BASE}/v1/chat/completions`, {
+      method: "POST",
+      headers: { "Content-Type": "application/json", Authorization: `Bearer ${apiKey}` },
+      body: JSON.stringify({
+        model,
+        stream: true,
+        max_tokens: 64,
+        reasoning_effort: "low",
+        messages: [{ role: "user", content: "Reply with the single word: hi" }],
+      }),
+    });
+    const raw = await drain(res);
+    expect(/Unsupported parameter.*thinking/i.test(raw), `${model} rejected 'thinking'`).toBe(false);
+    expect(res.status, `${model} bad status ${res.status}`).toBeLessThan(400);
+  }, 90000);
+});