Discover, Compare, and Track Every OpenClaw Variant

DeepSeek-V3 is a 671B MoE model (37B active) that matches top proprietary models at drastically lower training cost. DeepSeek-R1 achieves reasoning on par with OpenAI o1 via pure RL without supervised fine-tuning.

Qwen spans 100+ open-weight models from 0.6B to 32B, plus MoE and multimodal variants. Includes Qwen-Agent framework with function calling, MCP, and code interpreter, plus a terminal coding agent.

ChatGLM is a family of bilingual Chinese/English LLMs from Tsinghua University spinoff Zhipu AI. GLM-4.5 is a 355B MoE model with multimodal variants including CogVLM and CogVideoX.

Kimi K2 is a 1T-parameter MoE model (32B active) with SOTA performance among non-thinking models. K2.5 adds native vision via MoonViT, enabling agentic tasks like replicating website user journeys from video.

ERNIE 4.5 is a native multimodal model processing text, images, and video. Baidu open-sourced 10 models in the series under Apache 2.0, including a 47B MoE model with only 3B active parameters.

Doubao 2.0 is ByteDance's agent-era AI that decomposes complex objectives into executable sub-tasks using advanced reasoning chains. Doubao-Seed-Code achieved SOTA on SWE-Bench-Verified.

MiniMax develops multimodal AI with a novel Lightning Attention mechanism. M2.5 achieves SOTA in coding, tool use, and agentic workflows. Served 236M+ users across 200+ countries via Talkie and Hailuo AI.

Yi series from Kai-Fu Lee's 01.AI includes models that outperformed LLaMA2-70B at a fraction of the size. Yi-34B is fully bilingual English/Chinese with commercially permissive licensing.

Spark X1.5 is a 293B MoE model (30B active) with 130+ language support. Built on iFlytek's decades of speech AI expertise with leading voice cloning and mathematical reasoning capabilities.

Runs each Skill in OS-level sandboxes (Apple Sandbox or Docker), directly addressing OpenClaw RCE risks. Best personal alternative for users prioritizing security.

Optimized for RISC-V and ARM embedded boards, capable of 24/7 operation on $10 devices. Features Whisper voice integration, making it ideal for physical AI assistants and IoT deployments.

Ultra-fast, ultra-light Claw variant written in Rust. Sub-10ms startup time and minimal footprint make it ideal for scripting, automation, and edge computing.

The absolute minimum. No dependencies, sub-1MB binary — ideal for the most resource-constrained environments.

Built for stability and long-term continuous operation in industrial environments. C++ core ensures predictable performance over extended uptime.

Delivers core agent functionality with 99% fewer lines of code than OpenClaw.

Multi-agent collaboration platform with Docker container isolation. Features hierarchical agent swarms, web UI, and comprehensive tool support. Best for teams needing collaborative AI agents with security isolation.

Self-improving AI agent with automatic skill creation from experience. Features 40+ built-in tools, 6 deployment backends (local, Docker, SSH, Modal, Daytona, Singularity), 5+ messaging platforms (Telegram, Discord, Slack, WhatsApp, Signal, Email, CLI), works with 200+ models via OpenRouter. Server-first design with sandboxed execution and container hardening. Gets smarter over time by auto-creating skills from experience.

The original and most widely deployed Claw AI agent. Features the richest ecosystem with 5,000+ official and community skills, covering everything from everyday tasks to automated code review.