AMD has released Instella-MoE, a new open-source Mixture-of-Experts language model developed using their own GPUs and software. The model is available in various forms, including pre-trained, mid-trained, and fine-tuned versions, with its training code and frameworks also publicly accessible. Instella-MoE, with 16 billion total parameters and 2.8 billion active parameters, demonstrates competitive performance, particularly outperforming Gemma-4-E4B in certain benchmarks. AI
IMPACT This release provides a competitive open-source alternative for AI development, potentially accelerating adoption of AMD's hardware and software ecosystem.
RANK_REASON AMD, a major hardware vendor, has released a new open-source AI model. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Mastodon — fosstodon.org →
- AMD
- FarSkip-Collective
- Gemma-4-E4B
- Instella-MoE
- Instinct MI300X
- Instinct MI325X
- Primus
- Qwen3.5-4B-Base
- ROCm
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
