Axion 1.0, a fully open agent foundation model, launched with weights, training data, code, and an arXiv paper. It uses a three-stage recipe — two SFT stages then reinforcement learning — without mid-training, DPO, or on-policy distillation, and reportedly matches leading commercial models on several agentic benchmarks. The complete data construction, filtering, verification, and training pipeline is released for reproduction.
IQuest-Q1, a sparse Mixture-of-Experts agentic foundation model with 320B total parameters (15B activated), was released as an open-source model for agentic CLI and coding systems. Training combined Stable SplitNorm, Muon, and muP pre-training with multi-scaffold post-training across SFT, domain-specialized RL, and MOPD. The release targets community research and development.
IQuest-Q1, a sparse Mixture-of-Experts agentic foundation model with 320B total parameters (15B activated), was released as an open-source model for agentic CLI and coding systems. Training combined Stable SplitNorm, Muon, and muP pre-training with multi-scaffold post-training across SFT, domain-specialized RL, and MOPD. The release targets community research and development.
XYZ AI Lab released XYZ-Aquila-mini, an open-weight Deep Search agent post-trained from Qwen3.6-35B-A3B via an AI4AI pipeline. The model is a thinking model designed for bounded-exploration search tasks, with human-defined constraints and AI-driven failure diagnosis.
XYZ AI Lab released XYZ-Aquila-mini, an open-weight Deep Search agent post-trained from Qwen3.6-35B-A3B via an AI4AI pipeline. The model is a thinking model designed for bounded-exploration search tasks, with human-defined constraints and AI-driven failure diagnosis.
Intern-S2-Preview is a series of scientific agentic foundation models designed for multimodal scientific understanding, reasoning, generation, and long-horizon tasks. The training pipeline begins with scientific multimodal pre-training over rendered scientific documents, interleaved image-text data, and diverse scientific corpora.
Intern-S2-Preview is a series of scientific agentic foundation models designed for multimodal scientific understanding, reasoning, generation, and long-horizon tasks. The training pipeline begins with scientific multimodal pre-training over rendered scientific documents, interleaved image-text data, and diverse scientific corpora.
Agent Lightning v1.0 introduces harnessed agentic RL, where the deploy-time harness directly participates in model post-training. The original Agent Lightning's disaggregated architecture, connecting arbitrary agents to RL training via an LLM endpoint proxy, has been adopted by frameworks like verl Uni-Agent, AReaL 2.0, slime, and Polar.
Agent Lightning v1.0 introduces harnessed agentic RL, where the deploy-time harness directly participates in model post-training. The original Agent Lightning's disaggregated architecture, connecting arbitrary agents to RL training via an LLM endpoint proxy, has been adopted by frameworks like verl Uni-Agent, AReaL 2.0, slime, and Polar.
An arXiv paper introduces Rufus-Air, an open and reproducible post-training recipe built on GLM-4.5-Air-Base (106B-A12B). It details an eight-stage pipeline spanning SFT, multiple RL stages, agent training, and RLHF. The recipe improves over the official GLM-4.5-Air release and is competitive with similarly sized open models, making it a practical reference for practitioners reproducing full post-training pipelines.
An arXiv paper introduces Rufus-Air, an open and reproducible post-training recipe built on GLM-4.5-Air-Base (106B-A12B). It details an eight-stage pipeline spanning SFT, multiple RL stages, agent training, and RLHF. The recipe improves over the official GLM-4.5-Air release and is competitive with similarly sized open models, making it a practical reference for practitioners reproducing full post-training pipelines.