DeepGrove has released Maple-Preview, an open-source 20B-A1B ternary-weight LLM focused on reasoning capabilities. The model demonstrates state-of-the-art performance for its weight class, even competing with larger models and solving complex problems. It achieves impressive inference speeds, running at over 200 tokens/sec on a Mac Mini M4, significantly outperforming models like Gemma, Qwen 3.5, and gpt-oss. AI
IMPACT Sets new SOTA on reasoning benchmarks for its weight class, potentially influencing future efficient model development.
RANK_REASON Frontier-lab model release with system card [lever_c_demoted from frontier_release: ic=1 ai=1.0]
Read on Hugging Face Trending Models →
- DeepGrove
- deepgrove/maple-preview
- Docker
- Gemma
- gpt-oss
- Hugging Face
- Mac Mini M4
- Qwen 3.5
- SGLang
- transformers
- vLLM
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
