Google DeepMind has released Gemma 4 12B, a unified encoder free multimodal model built for running agentic AI locally on laptops. 🔥
- 12B parameter model that runs on laptops with 16GB memory
- Encoder free architecture for native image and audio processing
- Performance close to the larger 26B MoE model
- Native audio support with raw audio token processing
- Multi Token Prediction for lower latency
- Open sourced under Apache 2.0
- You can try here LM Studio, Ollama, Google AI Edge Gallery App, the Google AI Edge Eloquent app and the LiteRT-LM CLI
- New Gemma Skills Repository for agentic workflows
We are so happy to announce our new model Aion 1.0 today!
Our team at AI Frontiers Lab at Microsoft Research had been cooking hard on this for quite a while.
Aion 1.0 is 14B model that can run locally with reasoning + tool calling capabilities. You can choose whatever agentic harness you like or make your own. Calls to the model never leaves your device and no one charges you for any tokens you use 🥳.
“Immunotherapies are possible today only because thousands of scientists, for more than 40 years, followed their curiosity to probe the immune system’s deep processes.
Without basic scientific research, supported by the kind of farsighted public investment that allows large-scale, undirected, curiosity-driven inquiry, the scientific pipeline will run dry.” — MIT President Sally Kornbluth https://t.co/eO5X98XJNU