One model handles text, images, and audio on your device.
| Proof | 20M+ downloads of previous version |
| Best for | On-device search and privacy |
| Available via | Open weights (Apache 2.0) |
| Try it | Search audio logs with a text query |
Key takeaways
- Why it matters: You can find specific moments in long recordings without uploading data to the cloud.
- Before you switch: Check if your hardware supports 740M parameter models for smooth performance.
Source: Google DeepMind