Apertus v1.1 0.5B Instruct

A 4B parameter instruction-tuned language model created via pre-training distillation from the Apertus-8B teacher model. Uses a dense transformer decoder with grouped-query attention and xIELU activations, trained on 1.7T tokens. Natively supports 1811 languages with quantized checkpoints available for mobile and edge deployment. Apache-2.0 licensed.

Apertus-v1.1-4B is a 4B parameter open-source base language model created via pre-training distillation from the Apertus-8B teacher model. It uses a dense transformer with grouped-query attention and xIELU activations, natively supports 1,811 languages, and is trained on 1.7T high-quality tokens. Designed for constrained hardware and further fine-tuning.

Apertus-v1.1-1.5B-Instruct is a 1.5B parameter instruction-tuned language model built on a dense transformer decoder with grouped-query attention and xIELU activation. Created via pre-training distillation from an 8B teacher model trained on 1.7T tokens. Natively supports 1811 languages and optimized for mobile and edge deployments. Apache 2.0.

Hy-MT2 is Tencent Hunyuan’s open-source multilingual translation model family for fast-thinking, instruction-following translation across 33 languages.

A latency-optimized reranker that refines retrieval results with instruction-following capabilities. Supports a 32K token context window, works across 31 languages, and outperforms larger open-source rerankers. Users can steer relevance scoring using natural language instructions appended to queries.

XBai-o4 is an open large language model focused on test-time scaling and reflective reasoning improvements.