InternScience
The lab behind Agents-A1 — an open-weight (Apache-2.0) agentic-model effort that ships its weights on Hugging Face and a tech report on arXiv (2606.30616). On the strength of its first release it reads as a science/agent-focused open-weight challenger: the model’s six domains lean toward scientific research and engineering (FrontierScience, SciCode), and it’s built on the Qwen3 lineage rather than from scratch — fitting the pattern of open labs building on a shared open base. Too little public detail yet to place its provider strategy beyond “open-weight, agentic, research-leaning.”