Overview
- Kunlun Wanwei announced two models, SkyClaw‑v1.0 and a lightweight SkyClaw‑v1.0‑lite, and made them available through the TianGong Skywork platform with a 2–4 week free trial for users to test the models on real tasks.
- SkyClaw is described as supporting million‑token context windows and is tuned for agent workflows such as complex tool calls, multi‑step task execution, code generation, file editing, interactive app building, and data analysis.
- The company says the models were trained with large‑scale mid‑training, synthetic supervised fine‑tuning tasks and end‑to‑end reinforcement learning to improve task completion and multi‑step planning.
- Kunlun Wanwei reports internal and public benchmark results that place SkyClaw ahead of several open‑source models and close to larger top‑tier models on some agent tasks, and it says pricing is under half that of certain competitors; those claims have not been independently verified in this coverage.
- By offering immediate platform access and free trials, the launch lowers the barrier for developers to try long‑context agent automation and may accelerate demand for independent evaluations and cross‑platform integration work.