Research
Axion 1.0 open agent model trained in three stages with full recipe released
Axion 1.0 uses two supervised fine-tuning stages followed by reinforcement learning, skipping mid-training, DPO and on-policy distillation. The authors say it is competitive with leading commercial models on several agentic benchmarks and release training code and data pipeline.
Every outlet that ran it 1 article
Hugging Face PapersResearchLead