---

SkyClaw-v1.0 is a high-performance agent model by Skywork AI, optimized for complex tool use, multi-turn agent workflows, and cost-sensitive production tasks. Available in two variants:

Model	Input (CNY/M)	Output (CNY/M)	Best For
SkyClaw-v1.0	0.5	4.0	Strongest agent performance
SkyClaw-v1.0-lite	0.3	2.0	Speed & cost-sensitive tasks

🎉 Free for a limited time: Both SkyClaw-v1.0 and SkyClaw-v1.0-lite are currently free to use.

Benchmarks

SkyClaw-v1.0 outperforms Minimax 2.7, DeepSeek V4 Flash, and Qwen 3.6 series across all major agent benchmarks, while approaching larger proprietary models on Claw-related tasks.

Showcase

The release page features locally rendered screenshots and short videos, showcasing real generated demos across UI applications and interactive games.

Static Previews

Flight & Travel	Instagram-style	Xiaohongshu-style

Interactive Demos

Bouncing Balls Play	Bingo Match Play	2048 Puzzle Play
Tetris Play	Super Mario Play	Airplane Battle Play
Chess Play	Texas Hold'em Play	Financial Terminal Open
Tank Roguelike Play	Slay the Spire (杀戮尖塔) Play

Citation

If you reference SkyClaw-v1.0 in your work, please use the following citation:

@misc{skyclaw2026,
  title={SkyClaw-v1.0: A Million-Context Agent Model at Ultra-Low Cost},
  author={Peiyu Wang and Min Zou and Liang Zeng and Weishen and Peng Cheng and Haoran Zhang and Yu Cheng and Yang Liu},
  year={2026},
  month={May},
  howpublished={\url{https://skyworkai.github.io/skyclaw/}},
  url={https://skyworkai.github.io/skyclaw/},
}

Corresponding authors: Yu Cheng, Yang Liu