Gemopus Models Offer Local AI Solutions with Google’s Gemma Base
- The Gemopus-4-26B-A4B model activates around 4 billion of its total 26 billion parameters during inference, optimizing performance on constrained hardware.
- Gemopus-4-E4B is designed for edge devices, achieving speeds of up to 120 tokens per second on MacBook Air M3/M4 without a GPU.
- Both models are built on Google’s open-source Gemma 4 and avoid aggressive Claude-style reasoning distillation to enhance stability.
- The models passed extensive tests, including long-context and core competence evaluations, with the E4B variant excelling in all tested areas.
Jackrong’s Gemopus models leverage Google’s Gemma technology to deliver high-performance AI capabilities locally on personal hardware. The focus on stability and quality over imitation makes them a robust choice for users seeking Opus-style reasoning without relying on external processing power.
Gemopus models provide an American alternative to Qwen-based solutions, offering efficient local AI processing with impressive test results across various benchmarks. Source
