SkyClaw-v1.0 - A High-Performance Agent Model Launched by Kunlun Wanwei's Tiangong AI

What is SkyClaw-v1.0

SkyClaw-v1.0 is a high-performance Agent model launched by Kunlun Wanwei'sTiangong AI, supporting a million-token long context and deeply adapted toOpenClaw、Hermes、Claude Code、Codexand other mainstream Agent frameworks. The model has been optimized through large-scale mid-train, synthetic task SFT, and end-to-end reinforcement learning, surpassingMinimax 2.7、DeepSeek V4 Flashand other mainstream open-source models on benchmarks such as PinchBench-v2 and Claw-Eval. The model is now available for trial on the TiangongSkyworkplatform, and is also open for API access through APIFree.

Main Features of SkyClaw-v1.0

• Million-Context Agent Reasoning: Supports a million-token context window, handling complex multi-step tasks and long-document analysis.
• Multi-Framework Compatibility: Natively adapted to mainstream Agent environments such as OpenClaw, Hermes, Nanobot, Claude Code, and Codex.
• Complex Tool Calling: Deeply optimized for high-frequency Agent actions such as file reading, code editing, retrieval, testing, and page observation.
• Full-Form Application Building: Can generate multi-page applications with production-grade layouts, real navigation flows, and complete interaction states.
• Immersive Game Development: Supports generating interactive web games with physics engines, collision detection, state management, and user input.
• Research-Oriented Data Analysis: Can organize information, arrange data, present visualizations, and generate web reports around open topics.
• Cost-Effective API: Provides an OpenAI-compatible streaming API through APIFree, supporting tool calling and multi-turn conversations.

How to Use SkyClaw-v1.0

• Web: Log in to Tiangong Skywork, select SkyClaw V1.0 from the model list, and directly enter task instructions to use it.
• API Access: Register an APIFree account to obtain an API Key, and integrate SkyClaw into existing applications or Agent frameworks through an OpenAI-compatible interface.
• Agent Framework Integration: Configure SkyClaw as the underlying model for frameworks such as Claude Code, Codex, Hermes, or OpenClaw, enabling file editing, test execution, and multi-turn iteration capabilities.

Core Advantages of SkyClaw-v1.0

• Leading Performance: Surpasses mainstream open-source models such as Minimax 2.7, DeepSeek V4 Flash, and Qwen3.6-27B in evaluations such as PinchBench-v2, Claw-Eval Pass^3, and Skywork-Claw-Bench.
• Close to Top-Tier Large Models: Performs close to larger-scale models such as DeepSeek V4 Pro, Claude Opus 4.6, and Qwen 3.6 Plus on OpenClaw-related tasks.
• Extreme Cost-Effectiveness: Priced at less than half of Minimax 2.7 and the Qwen 3.6 series models, supporting large-scale high-frequency calls.
• Real Workflow Orientation: Optimized for continuous execution, error recovery, and multi-turn iteration in Agent frameworks.

Project Address of SkyClaw-v1.0

• Project Official Website：https://skyworkai.github.io/skyclaw/

Comparison of SkyClaw-v1.0 with Similar Competitors

Application scenarios of SkyClaw-v1.0

• Front-end application development: Rapidly generate production-grade interface prototypes for travel management, social applications, content communities, map navigation, and more.
• Interactive game creation: Develop web games such as physics simulation pinball, chess, airplane shooting, Texas Hold'em, and Roguelike card games.
• Data report generation: Create research-oriented web pages and PPTs such as financial news terminals, in-depth analysis of the Magnificent Seven U.S. stocks, and digital human broadcasts.
• Code Agent integration: Serve as an underlying model integrated into enterprise-level development workflows such as code review, feature development, and automated testing.
• Internal enterprise tools: Integrate via API into self-developed workflow systems, automated task platforms, and IM bots.