Atria Dawn Preview

Shanghai AI Laboratory's first Atria release: a "foundation agentic language model" for research and engineering workflows, built by post-training Z.ai's 744B-parameter GLM-5.2 MoE (the config.json is GLM-5.2's: 78 layers, 256 routed experts plus one shared, 8 active per token). The model card frames it around four kinds of agentic work, discovery, creation, delivery, and cybersecurity, and the accompanying 143-author paper describes a Verifiable Experience Pipeline that ties tool-mediated interactions to executable environments and externally verified outcomes. Released 11 September 2026 under MIT with a 256K context and an FP8 sibling; an Ascend W8A8 build followed on 15 September.

The paper reports that Atria Dawn Preview is competitive with frontier agents across 16 benchmarks and posts the highest reported score on five of them. From the card's table: DeepSearchQA 96.0, BrowseComp 92.5, BFCL v4 77.0, and AutomationBench 53.8 lead the compared column (DeepSeek V4 Pro 0813, Kimi K3, Qwen3.8 Max, GLM-5.3, GPT-5.6 Sol, Claude Opus 5), while SWE-bench Pro 59.6 and Terminal-Bench 2.1 78.3 trail the coding leaders. The paper also treats the model's own development as a case study in human–AI collaboration, analysing 769 task records from the R&D process. Not yet scored by Artificial Analysis.