Switch to fp16 ONNX model (87MB → 44MB)

No frontend code changes needed — onnxruntime handles fp16 internally
with float32 I/O. Max logit diff vs fp32 is 0.04 (negligible). Bump
model version to '3' to force re-download of cached models.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Author: BasedestBerry

public/maia3/maia3_simplified.onnx

@@ -120,7 +120,7 @@ export const DownloadModelModal: React.FC<Props> = ({
               </li>
             </ul>
             <p className="text-white/80">
-              Maia 3 runs entirely on your device and requires a one-time 87mb
+              Maia 3 runs entirely on your device and requires a one-time 44mb
               download.
             </p>
           </div>
@@ -175,7 +175,7 @@ export const DownloadModelModal: React.FC<Props> = ({
                 )}
                 <p className="font-medium">
                   {isDownloading ? 'Downloading...' : 'Download Maia 3'}{' '}
-                  <span className="text-xs text-white/70">(87mb)</span>
+                  <span className="text-xs text-white/70">(44mb)</span>
                 </p>
               </div>
             </div>

src/components/Common/DownloadModelModal.tsx

@@ -37,7 +37,7 @@ export const MaiaEngineContextProvider: React.FC<{ children: ReactNode }> = ({
       model:
         process.env.NEXT_PUBLIC_MAIA_MODEL_URL ??
         '/maia3/maia3_simplified.onnx',
-      modelVersion: process.env.NEXT_PUBLIC_MAIA_MODEL_VERSION ?? '2',
+      modelVersion: process.env.NEXT_PUBLIC_MAIA_MODEL_VERSION ?? '3',
       setStatus: setStatus,
       setProgress: setProgress,
       setError: setError,